JP4409421B2 - Memory device and access control method thereof - Google Patents

Description

  The present invention relates to a memory device having characteristics useful for, for example, digital processing of moving images and an access control method thereof.

  The need for this type of treatment is increasing. For example, many television receivers use digital image processing technology, and personal computers and workstations are often used for displaying images. In addition, moving images are increasingly exchanged between these different media, and conversion between different formats is required.

  Many digital image processes are performed on rectangular blocks of pixels. Typical examples are spatial and temporal filtering for noise reduction, effect processing and format conversion. Another example is motion prediction for image compression. When processing a moving image in real time, it is necessary to search a block of pixels at high speed. For example, when each new pixel of the video is received and stored, the next new pixel is received for the pixel block in which the new pixel occurred and the corresponding pixel block from several previous image frames or fields. It may need to be read before.

  Conventional DRAM is too slow for this task. This is because it is necessary to input a new address which takes time for every access for reading and writing.

  Conventional dual-port DRAM (also called video RAM, VRAM) allows high-speed serial readout of the entire row of pixels, eg, all pixels on a horizontal scan line on the screen. However, this feature is not useful for rectangular pixel blocks.

  Synchronous DRAM (SDRAM) and synchronous graphics RAM (SGRAM) allow burst access to smaller groups of pixels, but require different address inputs for reading and writing. This is inconvenient when both new reading and writing require both reading and writing. SDRAM and SGRAM also do not support the burst length that is most often required for digital filtering.

  In addition, any of these memories is difficult to make a cascade connection that allows access to pixel blocks in several frames or fields.

  Since the performance of a conventional RAM is insufficient, a FIFO memory is used to store a field or frame, and a rectangular pixel block is accessed using an ASIC (application specific integrated circuit) equipped with a line memory. For example, an ASIC comprising 21 line memories each consisting of SRAM (static RAM) and storing 1024 words (1 word = 8 bits) was used in a digital television receiver. SRAM memory cells are large and therefore line memory takes up a lot of space in the ASIC and space for the image processing circuit is reduced. In addition, since the SRAM operates as a shift register, a large current is consumed, and the cost of the ASIC increases.

  Details will be described later.

  An object of the present invention is to provide a memory device that combines the functions of a field or frame memory and the functions of a plurality of line memories.

  It is another object of the present invention to provide a memory device capable of performing burst read access and single write access by inputting a single row-column address.

  It is still another object of the present invention to provide a memory device capable of performing burst read access and single write access using a column address previously input by inputting a single row address.

  Still another object of the present invention is to provide a memory device suitable for slave connection.

  Yet another object of the present invention is to store pixel data for multiple fields or frames of a moving image and to output multiple pixel data from each field or frame in a single combined burst. An object of the present invention is to provide a memory device.

The memory device of the present invention includes:
A memory device having two memory banks for receiving a row and column address signal, input data, and an external control signal in synchronization with a clock signal,
A control signal generator ( 68 );
A data input unit (14) for receiving the input data;
A data output unit (16) for outputting output data;
A bank bus switch (56) and a transfer register (62) coupled to the bank bus switch and shared by all of the two memory banks;
Each memory bank (A, B) has a plurality of word lines,
A row decoder (6A, 6B) for activating a word line selected by the received row address signal among the word lines;
A main memory array (2A, 2B);
A column address generator (12A, 12B);
A main column decoder (8A, 8B);
A sub memory array (4A, 4B);
A sub-column decoder (10A, 10B);
Internal data buses (24A, 24B) and
The bank bus switch (56) couples the internal data bus (24A, 24B) to the data input unit (14) and the data output unit (16),
The main memory array (2A, 2B) has a plurality of memory cells arranged to form mutually intersecting rows and columns, and the word line is coupled to each row of the memory cells;
The column address generator (12A, 12B) generates a series of column addresses from a single received column address signal, each of the column addresses having an upper portion and a lower portion;
The main column decoder (8A, 8B) is coupled to the main memory array and the column address generator, decodes the column address, and transfers a column of the corresponding memory cell in the main memory array to the internal data bus ( 24A, 24B)
The sub memory array (4A, 4B) has a plurality of memory cells arranged to form mutually intersecting rows and columns, and the word line is also connected to each column of memory cells in the sub memory array. The number of columns in the secondary memory array is less than the number of columns in the main memory array;
The sub column decoder (10A, 10B) is coupled to the sub memory array and the column address generator, decodes a lower part of the column address, and selects a corresponding column of memory cells in the sub memory array as the internal memory. Coupled to the data bus,
The control signal generator (68)
Coupled to the column address generator, receives the external control signal, and based on this, enables the main column decoder and the sub column decoder, and generates an internal control signal for controlling the data input unit and the data output unit. And this
Input new input data into the data input unit (14),
Data stored in a plurality of columns of the main memory array of one of the two memory banks is read, and the internal data bus (24A ) and the bank bus switch ( 56) , old data read from one column of the plurality of columns transferred in this way is temporarily stored in the transfer register (62), and the data input unit ( The new input data input in 14) and the new data read from columns other than the one of the transferred plurality of columns are output through the data output unit (16). Let
Data stored in a plurality of columns of the main memory array ( 2B) of the other memory bank of the two memory banks is transferred to the internal data bus ( 24B) of the other memory bank, the bank bus switch (56) and between which Ru is outputted through the output unit (16), a new input data input to said data input unit (14), the internal data bus of one of the memory banks described above (24A) The old data stored in the transfer register (62) is transferred from the transfer register (62) to the one column of the main memory array (2A) of the one memory bank. Transferred to the secondary memory array (4A),
The old data stored in the transfer register (62) is output via the data output unit (16), and the old data stored in the sub memory array (4A ) of the one memory bank is output. the internal data bus (24 a), Ru is outputted through the bank bus switch (56) and said data output unit (16).

A method for controlling access to a memory array of a memory device of the present invention includes:
Two banks are provided in the memory device,
Each bank has a main memory array (2A, 2B) and a sub memory array (4A, 4B),
Different banks have different rows of memory cells,
The memory array of each bank has rows and columns of memory cells intersecting each other;
The secondary memory array of each bank has the same number of rows as the main memory array and fewer columns than the main memory array;
(A) receiving a row address signal and activating corresponding rows in the main memory array ( 2A ) and the sub memory array ( 4A ) in the first bank of the two banks;
(A ′) receiving a row address signal and activating corresponding rows in the main memory array ( 2B ) and the sub memory array ( 4B ) in the second bank of the two banks;
(B) generating, within the memory device, a first series of column addresses designating consecutive columns in the main memory array ( 2A ) in the first bank;
(C) inputting new input data into the first bank;
(D) Data is read from the memory cell arranged at the intersection of the row in the first bank activated in step (a) and the column specified in step (b). Of the data , old data read from the one column in the main memory array is transferred to the transfer register, and the transferred old data is temporarily stored in the transfer register and input to the data input unit. Outputting new input data and new data read from a column other than one column in the main memory array ;
(E) generating a second series of column addresses specifying memory cells in successive columns in the main memory array (2B) in the second bank;
(F) Read and output data from the memory cell arranged at the intersection of the row in the second bank activated in the step (a ′) and the column designated in the step (e) In the meantime, new input data input to the data input unit is stored in the one column in the main memory array in the first bank, and stored in the transfer register (62) in the step (d). Transferring stored old data to the sub-memory array (4A) in the first bank;
(G) generating, in the memory device, a third series of column addresses designating consecutive columns in the sub-memory array ( 4A ) in the first bank;
(H) The old data stored in the transfer register (62) is output via the data output unit, and the row in the first bank activated in the step (a) and the step ( g) reading out old data from the memory cell arranged at the intersection of the designated column and outputting the read out old data.

  As described above, according to the present invention, a memory device that combines the function of a field or frame memory and the functions of a plurality of line memories can be obtained. Further, a memory device capable of performing burst read access and single write access by inputting a single row-column address is obtained. Further, the input of a single row address provides a memory device that can perform burst read access and single write access using previously input column addresses. Furthermore, a memory device suitable for slave connection can be obtained. Furthermore, it is possible to store pixel data for a plurality of fields or frames of a moving image, and to obtain a memory device capable of outputting a plurality of pixel data from each field or frame in a single combination burst. .

  Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, before that, general things about moving image processing will be explained.

  One digital image processing is filtering. This corrects the value of a pixel by the values of surrounding pixels. In FIG. 1, an example of such filtering processing is as follows: the values of pixels indicated by hatched circles on the nth scanning line are represented on the (n−1) th scanning line, the nth scanning line, and the (n + 1) th scanning line. The correction is performed using adjacent pixels indicated by non-hatched circles.

  Such filtering may be performed in the time dimension direction. FIG. 2 shows five consecutive fields of the moving image. These five fields include odd-numbered fields Fm-2, Fm, Fm + 2, and even-numbered fields Fm-1, Fm + 1 interlaced therewith. The symbol Δt represents a field period, for example 1/60 or 1/50 second. Spatial and temporal filtering processes result in filtered pixel values at the positions indicated by hatching in the field Fm, depending on the values of all the pixels in the five groups DTm−2 to DTm + 2.

  FIG. 3 generalizes the process of accessing a 5 × 5 pixel block in five consecutive fields, where the spatial and temporal filtering is fast on pixels that are separated from each other by three types of delays. Indicates that access is required. The three kinds of delays are a bit delay (W) corresponding to an interval between consecutive pixels on the same line, a line delay (L) corresponding to an interval between consecutive lines, and an interval between fields. Is the field delay (t) corresponding to.

  The term bit delay is used because data bits of the same pixel are often accessed in parallel at the same time. This term does not imply that a pixel is represented by a single bit. A pixel is typically represented by 8 bits for a monochrome image and a larger number of bits for a color image.

  FIG. 4 shows a conventional system for accessing the pixel values shown in FIG. Here, it is assumed that one pixel is represented by 8 bits. At the left end, five cascaded FIFOs are shown. Each stores pixel data for one field. These field memories are dual port memory devices, where the output port of one memory is coupled to the input port of the next memory. Each pixel exists in each field memory for one field period Δt. Input and output are simultaneously performed in synchronization with the same clock signal (CLK). For example, the first field memory F1 receives the new pixel Da5 and at the same time passes the pixel Db5 at the same position in the previous field to the second field memory F2. The field memory thus has the field delay (t) shown in FIG.

  The line delay (L) is given by a plurality of line memories L11 to L54. Each pixel supplied to the first field memory F1 is also supplied to the first line memory L11. The line memory L11 is cascade-connected to the line memories L12, L13, and L14. The pixels are shifted through this slave connection in the same way as in the shift register, and each line memory spends one horizontal scanning period. Accordingly, the output of the line memory L11 is delayed by one line, and the output of the line memory L12 is delayed by two lines. Other line memories also introduce similar delays for the inputs of field memories F2-F5.

  The bit delay (W) is given by a sequence of D-type flip-flops DFF. Each pixel input to the field memory or output from the line memory is also supplied to the cascade connection of four D-type flip-flops. The outputs of these four flip-flops, together with the field and line memory inputs and outputs, allow simultaneous read access to all pixels of each 5 × 5 block of the five consecutive fields shown in FIG. To do.

  The line memory and D-type flip-flop of FIG. 4 are conventionally formed in an ASIC. However, as described above, the line memory is composed of SRAM, which requires a large space, consumes a large amount of current, and makes the device expensive.

  5 and 6 show two ways to synchronize block access to pixel input. Both of these figures show the pixels stored in one field. Hatched circles indicate new pixels belonging to the currently received field, and unhatched circles indicate pixels belonging to the previous field. A frame labeled “Read” indicates the position of a 5 × 5 pixel block used for filtering. The arrow marked “write” indicates the position (Dn) of the pixel currently being received.

  In FIG. 5, when a new pixel is written at position Dn in the field, the 5 × 5 pixel block of the previous field is being processed. The horizontal scan line containing Dn is the oldest scan line in this block. The pixel data is passed through a D-type flip-flop connected in cascade as shown in FIG. Therefore, it is sufficient to load the data from the five positions indicated by the bold circles Dn to Dn + 4 into the D flip-flop before writing a new pixel at the position Dn. All of these data are held in fields.

  FIG. 6 is a view similar to FIG. However, the horizontal scanning line including Dn is the newest scanning line in the block. Pixels Dn-1 to Dn-4 have already been overwritten, and Dn is about to be overwritten. In this case, it is necessary to add a means for storing the overwritten pixel. If the filtering operation requires both the current field and the previous field, pixel values must be output from positions Dn to Dn-4 in both fields.

  7 and 8 are enlarged views of a part of FIG. 7 indicates the field memory F1 and the line memories L21 to L24. These enable read access to the pixels Dn to Dn + 4 of FIG. A dotted frame in FIG. 8 indicates the field memory F1 and the line memories L11 to L24. These enable read access to the pixels Dn to Dn-4 other than the pixel Da5 in the current field and the previous field in FIG.

  The present invention provides a single memory device having all the functions of the portion surrounded by the dotted line in FIG. 7 or the portion surrounded by the dotted line in FIG. The memory device is further suitable for the cascade connection of the field memories shown in FIGS.

First Embodiment A memory device according to a first embodiment is shown in a block diagram of FIG. This has a function of a portion surrounded by a dotted line in FIG. As an example of the first invention of the present application, the memory device includes a main memory array 2, a sub memory array 4, a row (X) decoder 6, a main column decoder (MYD) 8, a sub column decoder (SYD) 10, a column ( Y) It has an address generator 12, a data input unit (DI) 14, and a data output unit (DO) 16. The data input unit 14 has at least one data input terminal DIN. The data output unit 16 has at least one output terminal DOUT.

  In the following description, the letters X and Y are used to represent rows and columns.

  The Y address generator 12 is coupled to the main Y decoder 8 by an upper address bus 18 and is coupled to the main Y decoder 8 and the sub Y decoder 10 by a lower address bus 20. The lower address bus is divided into two parts by an address bus switch 22. The first part of the lower address bus is connected to the main Y decoder 8. The second part of the lower address bus is connected to the secondary Y decoder 10 and can be separated from the Y address generator 12 by the address bus switch 22.

  The data input unit 14 and the data output unit 16 are connected to the main memory array 2 and the sub memory array 4 by an internal data bus 24. The internal data bus 24 is divided into two parts by a data bus switch 26, and one end is connected to a write amplifier (WAMP) 28. The two parts of the internal data bus 24 are called a main data bus (MDB) and a sub data bus (SDB).

Each of the circles in the main memory array 2 and the sub memory array 4 represents a group of memory cells having the same XY (row-column) address. In image processing applications, each group stores data for one pixel. In a moving image scanned by successive horizontal scanning lines, pixels having the same column address (Y address) are arranged on the same scanning line. Thus, the columns extend horizontally in the drawing and the rows extend vertically . The memory cells in each row are connected to a word line (WL). The word line is common to the main memory array 2 and the sub memory array 4.

  FIG. 10 shows the configuration of the first embodiment in more detail. The X decoder 6 is connected to an X address generator 29, and the X address generator 29 receives a row address XAD inputted from the outside. The Y address generator 12 receives a column address YAD input from the outside, and includes a down counter (DC) 30 and a Y address output circuit (YADOUT) 32. The down counter 30 operates as a column address counter and generates a series of column addresses (Y addresses) that are sequentially reduced. The data input unit 14 has an internal latch (LT) 34 for holding input data. The data output unit 16 has an output amplifier (OAMP) 36 for amplifying output data. The main and sub Y decoders MYD and SYD have a plurality of AND gates 38 connected to the upper address bus 18 and the lower address bus 20. An address holding latch (HLD) 40 is provided for each address line in the lower address bus 20.

  The AND gates 38 in the main Y decoder are given symbols And1 to Andn, and the AND gates 38 in the secondary Y decoder are given Ad1 to Ado. The letters “n” and “o” represent the number of columns in the main memory array 2 and the sub memory array 4, respectively. The sub memory array 4 has fewer columns than the main memory array 2 (o <n). The AND gate 38 generates a column selection signal, which is represented by Y1 to Yn in the main Y decoder and Ys1 to Yso in the sub Y decoder, respectively.

  Each portion of the data bus has a pair of complementary bus lines Dm and Dm / in the main data bus and a pair of complementary bus lines Ds and Ds / in the sub data bus. The bus lines Dm and Ds are coupled to each other via a transmission gate formed of a pair of reverse channel transistors in the data bus switch 26. Similarly, the bus lines Dm / and Ds / are coupled to each other via a transmission gate in the data bus switch 26, and the lower address signal line is opened and closed by the transmission gate in the address bus switch 22. These transmission gates are controlled by a sub memory enable (SME) signal. The SME signal is directly supplied to the negative channel transistor of each transmission gate, and is supplied to the positive channel transistor of the transmission gate after being inverted by the inverter 42. The SME signal is also supplied to the address holding latch 40 via the inverter 44 for control.

  The main memory array 2 has a plurality of pairs of complementary bit lines BL1 and BL1 / to BLn and BLn /. These extend in the column direction, that is, in the direction perpendicular to the word lines WL1 to WLn. Each dynamic memory cell Nij includes a transistor and a capacitor, and is connected to a word line and a bit line as shown in the figure. A sense amplifier (SA) is connected to each pair of complementary bit lines. The complementary bit lines in the main memory array 2 are connected to complementary data bus lines Dm and Dm / by a pair of transfer transistors 46. Transfer transistor 46 is driven by a column select signal from one of AND gates 38 in the main decoder.

  Similarly, the sub memory array 4 includes a plurality of pairs of complementary bit lines BLs1 and BLs1 / to BLso and BLso /, a memory cell Nsij, a sense amplifier SA, and a transfer transistor 46. Both sense amplifiers in the main memory array 2 and the sub memory array 4 are activated by a single sense amplifier control signal PSA.

  For simplicity, FIG. 10 shows only one memory cell at each XY address. As described above, a group of memory cells is generally arranged at each XY address. The data bus has a plurality of complementary pairs of bus lines. In each column, there are the same number of complementary pairs of bit lines and the same number of sense amplifiers, and the data input unit 14 and the data output unit 16 have the same number of input and output terminals.

  Among the members of the first embodiment, a memory control signal generator is not shown in FIGS. 9 and 10. This generates the control signal shown in FIG. In addition to the SME and PSA control signals already mentioned, these control signals include a control signal R / W. The control signal PYE includes a control signal PYE supplied to the AND gate in the secondary Y decoder and a control signal R / W supplied to the Y address output circuit 32. The memory control signal generator is illustrated in a later embodiment.

  Next, the operation of the first embodiment will be described.

  The memory control signal generator can be programmed to provide control in various operating modes. The first mode enables access to the pixel data shown in FIG.

This mode will be described with reference to FIG. In accordance with the convention for FIG. 6, the hatched circles in FIG. 9 indicate groups of memory cells that store the pixels in the current field, and the unhatched circles indicate groups of memory cells that store data in the previous field.

  In this mode, the X address is received, decoded by the X decoder 6, and the corresponding word line WL is activated. All memory cells (in the main memory array 2 and sub memory array 4) connected to the word line are thereby connected to the corresponding bit lines, and the data stored in these memory cells is amplified by a sense amplifier. Is done.

  Next, the first Y address is generated by the Y address generator 12, decoded by the main Y decoder 8, and the corresponding column in the main memory array 2 is selected. Memory cells located at the intersections of this column and the row of word lines WL are connected to the data bus 24 and their data are transferred to the data output unit 16. This is indicated by a thick line (A).

  At the same time, these data are transferred from the main memory array 2 to the sub memory array 4. This is indicated by the arrow (B). This transfer is performed via the internal data bus 24. At this time, the main data bus and the sub data bus are connected by the data bus switch 26. Data is amplified by write amplifier 28. As soon as the write amplifier 28 obtains data, the data bus switch 26 is opened to separate the main data bus and the sub data bus. At substantially the same time, the secondary Y decoder 10 decodes the lower address bits of the Y address and selects the corresponding column in the secondary memory array 4. The data amplified by the write amplifier 28 is written into a memory cell located at the intersection of this column in the sub memory array 4 and the word line WL.

  Next, the down counter 30 in the Y address generator 12 generates Y addresses that sequentially decrease, and the main Y decoder 8 sequentially selects the columns in the main memory array 2 as indicated by arrows (C). . Data in the memory cells on the word line WL in these columns is sequentially transferred to the data output unit 16. The data output unit 16 outputs the transferred data from the data output terminal (DOUT) with a certain latency.

  During the transfer of these data (C), the data bus switch 26 remains open, so that the transfer does not affect the secondary memory array 4. Since the address bus switch 22 is also open, the secondary Y decoder 10 does not receive the sequentially changing Y address and continues to receive the first Y address from the address holding latch 40. Therefore, while the data (C) is sequentially transferred from the main memory array 2 to the data output unit 16, the data (B) write operation to the selected memory cell in the sub memory array 4 is continued.

  In the operation shown in FIG. 9, the transfer of data from the main memory array 2 to the data output unit 16 ends when three columns are accessed. By this time, the operation of writing the first transferred data to the sub memory array 4 has been completed.

  Next, the address bus switch 22 and the data bus switch 26 are closed and the Y address generator 12 again generates the same series of three Y addresses. Data stored in the corresponding memory cell (D) on the word line WL in the sub memory array 4 is transferred to the data output unit 16 via the data bus 24, and is output with the above-described latency. Data from the main memory array 2 and the sub memory array 4 is thus output as a single continuous serial burst.

  On the other hand, at the data input terminal (DIN), new input data is received and held in the latch 34 in the data input unit 14. After the transfer of the data (D) from the sub memory array 4 to the data output unit 16 is completed, the Y address generator 12 generates the start Y address again, and new input data is transferred from the input unit 14 to the main memory array 2. (E) and written to the same memory cell that was read at the beginning of the burst.

  One reason for writing new data at the end of a burst is that this allows read access to be performed continuously without interruption. If new data is to be written in the middle of a burst, extra control is required for read-write-read switching, whereas if read-write switching is performed only once, the operation can be performed. Become faster.

  Another reason for writing new data at the end of the burst is the latency of the data output unit 16. This allows new data to be transferred via the data bus 24 in the background (background) while the data output unit 16 is still outputting the data read from the sub memory array 4. To do.

  The relationship between the above operation and FIG. 8 is as follows. The operation (A) in FIG. 9 corresponds to outputting Db5 from the field memory F1 to the field memory F2 in FIG. The operation (C) in FIG. 9 corresponds to outputting data Da4 to Da1 from the line memories L11 to L14 in FIG. The operation (D) in FIG. 9 corresponds to outputting data Db5 to Db1 from the field memory F1 and the line memories L21 to L24 in FIG. The operation (E) of FIG. 9 corresponds to inputting new data Da5 to the field memory F1. Thus, the first embodiment has the functions of the field memory F1 and the line memories L11 to L24 in FIG.

  When the memory device of the first embodiment is used for digital processing of moving images, the ASIC that controls the memory device does not require eight line memories L11 to L24 that require thousands of SRAM memory cells, but instead is short (for example, It is sufficient to store a burst of data read from the memory device by means of a 9-stage shift register. An example of this type of shift register will be described later with reference to FIG. In this way, the size, price and current consumption of the ASIC are significantly reduced.

  Note that the output of the data Db5 occurs twice during the above operation. One time is from the main memory array 2 at the start of the burst, and the other time is from the sub memory array 4 in the middle of the burst. This dual output can be used as described later when the memory devices are cascade-connected.

  Several aspects of the above operation of the first embodiment will be described in more detail.

  The operation of the first embodiment is performed in synchronization with the synchronization signal (CLK) shown in FIG. Furthermore, the first embodiment receives the following external control signals. That is, chip selection (CS /), row address strobe (RAS /), column address strobe (CAS /), and write enable (WE /). A slash “/” in the signal name indicates that the signal is at a low level when activated. The first embodiment further includes an external address bus (ADD). The address input and other input and output signals are synchronized with the rising edge of the clock signal (CLK).

  FIG. 11 shows operations of reading data from the main memory array 2 and transferring data from the main memory array 2 to the sub memory array 4 (operations (A), (B), and (C) in FIG. 9). The waveforms of the external control signal and the internal control signal are shown. The internal control signal is generated according to the operation mode information programmed in the control signal generator in the control signal generator mentioned above based on the external control signal.

  At time t1, CS / and RAS / are at a low level, and CAS / and WE / are at a high level. This combination of control signals indicates that the address on the address bus (ADD) is a row address (Xi: supplied to the X decoder 6 by the X address generator 29). The X decoder 6 activates the corresponding word line WLi (WL3 in FIG. 11), and the capacitor in the memory cell on this word line is replaced with the bit lines BLj and BLj / (j = 1 to n) and the sub-cells in the main memory array 2. The bit lines BLsj and BLsj / (j = 1 to o) in the memory array 4 are connected. The data stored in these capacitors appears as a small potential difference between each complementary pair of bit lines. Next, a sense amplifier control signal PSA (not shown) activates the sense amplifier, and the sense amplifier amplifies the potential difference on the bit lines in both memory arrays and converts it to a signal having a full power supply voltage.

  At time t2, CS /, CAS /, and WE / are at a low level, and RAS / is at a high level. This combination of control signals indicates that the address on the address bus (ADD) is a column address (Yj: loaded into the down counter 30 in the Y address generator 12). This combination of signals also indicates the presence of input data at the input data terminal (DIN). Input data is latched in a latch 34 in the data input unit 14. However, it is not immediately written into the main memory array 2.

  Assume that the four lower bits of the Y address Yj, and therefore the values of the four lower bits YA3, YA2, YA1, YA0 of the start address output by the down counter 30 are “0110” as shown. An AND gate 38 in the main Y decoder 8 decodes this start address. A column select signal (also referred to as Yj) output from only one of these AND gates 38 is high during the 1/2 clock cycle starting at time t2, and during that 1/2 clock cycle. The bit lines BLj and BLj / are connected to the main data bus lines Dm and Dm /. Therefore, the potentials on the bit lines BLj and BLj / are transferred as data D1 to the data bus lines Dm and Dm / via the transfer transistor 46.

  The enable signal SME is at a high level at this time, and the main data bus lines Dm and Dm / are connected to the sub data bus lines Ds and Ds / via the data bus switch 26. Therefore, the data D1 is also transferred to the sub data bus lines Ds and Ds /.

  As indicated by the letter L, the R / W control signal is at a low level during the read operation.

  The PYE control signal is low for a 1/2 clock cycle starting at time t2, during which the secondary Y decoder 10 is disabled (inactive). The column selection signals output from all the AND gates 38 are kept at a low level during this period. As a result, the write amplifier 28 is given time to latch and internally amplify the potential (D1) on the sub data bus lines Ds and Ds /.

  Write amplifier 28 has an enable signal, which is not shown for simplicity. The main Y decoder enable signal is also not shown. This main Y decoder enable signal is at a high level in the operation shown in FIG. 11 and enables the main Y decoder 8. This main Y decoder enable signal may be supplied to the AND gate 38 in the main Y decoder 8 in the same manner as PYE is supplied to the AND gate 38 in the secondary Y decoder 10. The Y decoder enable signal may be supplied to the Y address output circuit 32 and then supplied from the Y address output circuit 32 to the AND gate 38 as an extra address bit.

  After 1/2 clock cycle has elapsed from time t2, PYE goes high. The sub-Y decoder 10 decodes the lower address bits, and the column selection signal Ysj output from the corresponding AND gate 38 in the sub-Y decoder 10 becomes high level, and the bit line pair BLsj and BLsj / is changed to the sub-data bus line Ds. And Ds /. The sub data bus lines Ds and Ds / are held at the potential of the data D1 by the write amplifier 28. Regardless of the previous potentials of the bit lines BLsj and BLsj /, the write amplifier 28 drives these bit lines to the potential of the data D1.

  After the elapse of one clock cycle from time t2, the value of the down counter 30 is decreased by 1, and the lower address bits are changed from “0110” to “0101” (read in the order of YA3 to YA0). However, the output of the Y address output circuit 32 remains “0110” until immediately before time t3. At this time, the output amplifier 36 in the data output unit 16 starts outputting the data D1 from the output terminal (DOUT).

  Shortly before time t3, the enable signal SME goes low, separates the sub data bus lines Ds and Ds / from the main data bus lines Dm and Dm /, and further inputs the lower address bits to the sub Y decoder 10. Stop. The address holding latch 40 continues to hold the lower bit value (“0110”) of the start address, the signal Ysj remains at the high level, and the data D1 of the bit lines BLsj and BLsj / is stored in the corresponding memory in the sub memory array 4. Written in the cell. Writing data D1 can continue for any period of time while other data is being read from main memory array 2, and write amplifier 28 takes sufficient time to charge or discharge the capacitor being written. Can be done.

  At time t3, the Y address output circuit 32 outputs the decreased Y address (lower order is “0101”). This address is decoded by the main Y decoder 8, and the column selection signal Yj-1 output by the AND gate 38 in the previous column becomes high level. Data D2 is read from the bit lines in this column and transmitted to the main data bus lines Dm and Dm /. The read latency from the input of the column address signal to the output of the first data of the burst is one clock cycle.

  In this way, the burst operation is continued. At time t4, the Y address output circuit 32 outputs a further reduced Y address (lower order is “0100”), and the transfer of data D3 from the main memory array 2 to the main data bus lines Dm and Dm / starts. Data D2 is still output from the terminal DOUT. Further explanation of this part of the burst is omitted.

  When reading of data from the main memory array 2 is completed, reading of data from the sub memory array 4 is performed in the same manner. At this time, the enable signals SME and PYE are both at a high level, and the main Y decoder enable signal (not shown) is at a low level. The down counter 30 restarts counting from the Y address Yj (the lower order is “0110”).

  The read latency of these operations is not limited to one clock cycle. The memory device may have a longer read latency and the read latency may be programmable.

  Details of the operation of writing new data to the main memory array 2 will be omitted here, and a later embodiment will be described.

  Next, the second operation mode of the first embodiment will be described. In this mode, reading of data from the sub memory array 4 is not automatically performed following reading of data from the main memory array 2. Instead, data is read from the sub memory array 4 in response to an external command. This mode is used when an external device needs to obtain data from the main memory array 2 and the sub memory array 4 at different times, or from one row in the main memory array 2 and another row in the sub memory array 4. Useful when you need to get.

  The reading of data from the main memory array 2 and the transfer of data from the main memory array 2 to the sub memory array 4 are shown in FIG. 11 and are performed in the same manner as described above.

  FIG. 12 shows reading of data from the sub memory array 4. During this operation, the SME signal is high and the R / W signal is low, as indicated by H and L, respectively.

  Prior to this operation, the control signal generator is programmed with an external command (not shown) that specifies access to the sub memory array 4. Accordingly, when the X address (Xi) is input at time t1, the enable signal PYE becomes high level.

  At time t2, CS / and CAS / are at a low level, and RAS / and WE / are at a high level. Therefore, even if the Y address (Yj) is received, new input data is not latched. This Y address (lower order is “0110”) is immediately output by the down counter 30 and the Y address output circuit 32. Since SME is at a high level, the lower bits of the Y address are transmitted to the secondary Y decoder 10 via the address bus switch 22. Since PYE is high, the secondary Y decoder 10 is enabled and decodes these lower bits. The column selection signal Ysj output by the corresponding AND gate 38 in the secondary Y decoder 10 becomes high level (for one clock cycle), and the memory cell on the word line WLi (WL3) of the corresponding column in the secondary memory array 4 The data Ds1 stored in is transferred to the sub data bus lines Ds and Ds /. Since the SME is at a high level, these data Ds1 are transferred to the main data bus lines Dm and Dm / through the data bus switch 26 and further to the data output unit 16.

  At time t3, the data Ds1 amplified by the output amplifier 36 is output from the output data terminal DOUT. At this time, the value of the down counter 30 decreases to the next lower address (ends with "0100"), the secondary Y decoder 10 activates the column selection signal Ysj-1, and the data Ds2 becomes the signal Ysj- 1 is transferred from the memory cell selected by 1 to the data bus line.

  The read operation is thus continued at times t4 and t5. Further explanation is omitted.

  As a modification, the first embodiment can be configured to always operate in the second mode. In this case, a separate address input is required for each burst access to the sub memory array 4. A separate address input can also be used for write access to the main memory array 2. Various other modes of operation are possible. The basic feature of the first embodiment is that it shares the same word line and lower Y address as the main memory array 2, and data is automatically transferred from the main memory array 2 when the main memory array 2 is accessed. This is the presence of the secondary memory array 4.

  As yet another modification, the timing can be determined so that the new data Da5 is input simultaneously with the first output of the old data Db5. Referring to FIG. 11 again, in this case, CS / and CAS / are set to low level at time t2, CS / and WE / are set to low level at time t3, and new data is input to input terminal DIN at time t3 (not time t2). To supply. Instead, at the output of the first data of each burst, the data input unit 14 operates the memory device to operate in a mode that automatically latches new data, even if no WE / signal is input separately. It can also be programmed. As shown in FIG. 8, this modification is convenient when the first embodiment (F1) is cascade-connected to another memory device (F2). This is because both memory devices F1 and F2 can simultaneously receive input data.

Second Embodiment Referring to FIG. 13, the second embodiment is similar to the first embodiment. The difference is the configuration of the data bus. The same or corresponding members as those in the first embodiment are denoted by the same reference numerals. Hereinafter, differences from the first embodiment will be described.

  In the first embodiment, the main and sub data buses are interconnected by the data bus switch 26. In the second embodiment, they are not interconnected. Instead, the main data bus lines Dm and Dm / and the sub data bus lines Ds and Ds / are independently connected to the output amplifier 36 in the data output unit 16. The output amplifier 36 plays the same role as the data bus switch 26 by transferring data from the main data bus to the sub data bus. The output amplifier 36 also amplifies data to be written to the sub memory array 4 and plays the role of the write amplifier 28 of the first embodiment. As a result, the data bus switch 26 and the write amplifier 28 can be omitted.

  The main data bus lines Dm and Dm / are connected to a latch 34 in the data input unit 14. The sub data bus lines Ds and Ds / are not connected to the data input unit 14.

  The second embodiment operates in the same manner as the first embodiment. However, with the omission of the data bus switch 26, the electric resistance of the transistor of the data bus switch 26 is lost, and access to the sub memory array 4 is further accelerated. Furthermore, the electric capacitance of each data bus line is reduced, which also contributes to speeding up. The second embodiment can therefore operate at a higher clock rate than the first embodiment.

  Since the data bus switch 26 and the write amplifier 28 are eliminated, the circuit configuration is simplified. This is a further advantage.

Third Embodiment Referring to FIG. 14, the third embodiment is similar to the first embodiment. However, the column (Y) address bus configuration is different. The same or corresponding members as those in the first embodiment are denoted by the same reference numerals. Hereinafter, differences from the first embodiment will be described.

  In the first embodiment, the lower address bus 20 is directly connected to the main Y decoder 8 and is connected to the secondary Y decoder 10 via the address bus switch 22. In the third embodiment, the lower address bus 20 is provided in duplicate, one set of lower address lines 47 is connected to the main Y decoder 8, and the other set of lower address lines 48 is connected to the sub Y decoder 10. It is connected. The address bus switch 22 and the address holding latch 40 of the first embodiment are omitted.

  A combination of the upper address bus 18 and the lower address line 47 constitutes a main address bus, which transmits the upper and lower portions of the Y address from the Y address generator 12 to the main Y decoder 8. The lower address line 48 constitutes a sub address bus and transmits the lower part of the Y address from the Y address generator 12 to the sub Y decoder 10.

  The third embodiment operates in the same manner as the first embodiment. However, the omission of the address bus switch 22 eliminates the electrical resistance of the transistors of the address bus switch 22 and speeds up access to the sub memory array 4. Furthermore, since the electric capacitance of the lower address bus line is reduced, this also contributes to speeding up of access. Therefore, the third embodiment can operate at a faster clock rate than the first embodiment.

  Since the address bus switch 22 and the address holding latch 40 are eliminated, the circuit configuration is simplified. This is a further advantage of the third embodiment.

Fourth Embodiment Referring to FIG. 15, the fourth embodiment is a combination of the features of the second embodiment and the third embodiment. The main data bus lines Dm and Dm / and the sub data bus lines Ds and Ds / are separately connected to the output amplifier 36 in the data output unit 16, and are separated from the main Y decoder 8 and the sub Y decoder 10 by separate lower address buses. Lines 47 and 48 are provided. The address bus switch 22, data bus switch 26, write amplifier 28, and address holding latch 40 of the first embodiment are all removed.

  The fourth embodiment combines the advantages of the second and third embodiments in that the circuit configuration is simple and the operation is fast.

  The operation of the first to fourth embodiments will be summarized with reference to FIG. Again, Da1 to Da5 represent the pixel data of the current field, and Db1 to Db5 represent the pixel data of the previous field. When new pixel data Da5 is received, data Db5 of the same pixel in the previous field is transferred from the main memory array 2 to the sub memory array 4. Data Db5, Da4 to Da1, and Db5 to Db1 are output as bursts from the main memory array 2 and the sub memory array 4, and new data Da1 is written to the main memory array 2. These data all have the same row address (X address) in the memory device.

  As indicated by the oval arrows, the sub memory array 4 is used cyclically. In the third and fourth embodiments, the main Y decoder 8 and the sub Y decoder 10 have separate address bus lines. As long as the sub memory array 4 is cyclically used, the address supplied to the sub Y decoder 10 is used. Need not be the same as the lower bits of the address supplied to the main Y decoder 8. In this case, the number of columns of the sub memory array 4 may not be 2, and may be 6, for example, as shown in FIG.

  In the above description, the data Db5 is output twice in the same burst. However, this is not always necessary. For example, new data Da5 may be transferred from the data input unit 14 to the data output unit 16, and Da5 may be output instead of Db5 at the beginning of the burst. Also in this case, the old data Db5 is transferred from the main memory array 2 to the sub memory array 4. Specific examples will be described in the following embodiments.

  In the above description, description of means for controlling the burst length and means for repeatedly loading the same Y address into the down counter 30 is omitted. Examples of these means will be described in the following embodiments. However, the present invention is not limited to the means shown in the following embodiments of the first to fourth embodiments. As described above, the first to fourth embodiments can be configured to operate by repeatedly inputting the same address from the outside. The burst length can also be configured to be controlled externally.

Fifth Embodiment Referring to FIG. 17, the fifth embodiment is similar to the first embodiment. The same or corresponding members are denoted by the same reference numerals. Unlike the first embodiment, three additional members are provided in the Y address generator 12. These additional members are an access counter (AC) 50, an address register (ADR) 52, and an address register output switch 54. The access counter 50 receives the clock signal CLK and the stop control signal PST, and outputs an address input control signal PAI to the address register output switch 54. The address register 52 receives and stores an externally input Y address YAD. Address register output switch 54 couples address register 52 to down counter 30.

  The control signal PST initializes a value for controlling the burst length in the access counter 50. This value may be a constant value preset in the access counter 50, and is a programmable value stored in a memory control signal generator (not shown), which is transmitted to the access counter 50 by the control signal PST itself. It may be a thing. The access counter 50 only needs to output the control signal PAI. Therefore, the access counter 50 can be constituted by a ring counter or a shift register, for example.

  Although not explicitly shown in the drawings, the fifth embodiment also allows the input data held in the latch 34 in the data input unit 14 to be transmitted via, for example, the main data bus lines Dm and Dm / or the data bus line. And a control signal for transfer to an output amplifier 36 in the data output unit 16 via a direct interconnection line bypassing. This transfer enables the input data to be output from the data output unit 16 without being stored in the main memory array 2 once.

  Referring to FIG. 18, the fifth embodiment receives the same CS /, RAS /, CAS /, WE / control signal as the previous embodiment. In addition, the address transfer control signal ADX / is received. The memory device uses this control signal ADX / to generate an X or Y address signal internally even if there is no input on the address bus ADD. When RAS / and ADX / are both low, the X address generator 29 generates a new X address by incrementing the previous X address. When CAS / and ADX / are both at a low level, the Y address held in the address register 52 is transferred to the down counter 30 via the address register output switch 54.

  In the operation shown in FIG. 18, the read latency is two clock cycles, unlike the previous embodiment. Following the input of the new X address Xi and Y address Yj and the new data Da5, the input data Da5 is transferred from the data input unit 14 to the data output unit 16 as described above. The input data Da5 is also held in the data input unit 14. Old data Db5 existing in a memory cell to which new data Da5 is to be written is transferred from the main memory array 2 to the sub memory array 4. In this case, the data output unit 16 is not latched. Thereafter, similarly to the previous embodiment, data Da4 to Da1 are read from main memory array 2, and data Db5 to Db1 are read from sub memory array 4.

  Therefore, the data output from the data output terminal (DOUT) is Da5, then Da4 to Da1, and then Db5 to Db1. Thus, five pixels from the current field and five corresponding pixels from the previous field are provided.

  In FIG. 18, an arrow X indicates reloading of the address Yj from the address register 52 to the down counter 30. This operation is controlled by the access counter 50 as follows. In the clock cycle when the column address strobe signal CAS / is at a low level, the internal control signal PST initializes the access counter 50 to a value “5”, for example. While the down counter 30 counts five consecutive Y addresses (Yj to Yj-4), the access counter 50 counts down from “5” to “0”. When the count value of the access counter 50 becomes zero, the access counter 50 activates the PAI control signal, and the address Yj held in the address register 52 is loaded into the down counter 30 again. Therefore, in the next clock cycle, the address register 52 starts counting down from Yj again.

  In this way, the down counter 30 easily generates a Y address for reading data from both the main memory array 2 and the sub memory array 4 without the second address input. During the reading of data from the secondary memory array 4, the access counter 50 also counts down from “5” to “0” again. Following the reading of data from the secondary memory array 4, the access counter 50 activates the control signal PAI again and loads the start address (Yj) into the down counter 30 once again. New input data Da5 held in the latch 34 in the data input 14 is transferred to the main memory array 2 and written to the memory cell previously occupied by the data Db5.

  After this burst access, the external control signal ADX / goes low twice. The first is with CS /, RAS /, and the second is with CS /, CAS /, WE /. With these commands, the X address generator 29 generates the next X address (Xi + 1), and the access counter 50 reloads the same Y address (Yj) into the down counter 30. Accordingly, five pixels on the right side of Da5 to Da1 and Db5 to Db1 of the current field and the previous field are output by the next burst.

  Thereafter, a new address can be generated by ADX / and a burst can be output in the same manner. Pixel data for the entire horizontal scan line can thus be received and stored in a memory device, and a burst output required for filtering can be provided. In addition, this can be realized by preparing only one X address and Y address at the beginning of the scanning line.

  FIG. 18 further shows the advantages of the fifth embodiment over the conventional SDRAM. The burst length of the fifth embodiment is limited (indirectly) only by the size of the sub memory array 4. On the other hand, the burst length in the conventional SDRAM is limited to specific values “1”, “2”, “4”, “8”. An unrestricted burst is possible, but in that case the burst must be stopped by external control. In a conventional SDRAM, for example, it is difficult to obtain 10 data read from 5 consecutive addresses in 2 fields as a continuous burst.

  The fifth embodiment plays the same role as the field memory F1 and the line memories L11 to L24 surrounded by a dotted line in FIG. Compared to the first to fourth embodiments, there are more functions in that an output of new data Da5 is included. However, when the memory devices are cascade-connected, this configuration is not as appropriate as the first to fourth embodiments. In FIG. 19, the memory device F1 is used alone, whereas in FIG. 8, the memory device F1 is cascade-connected to the memory device F2.

  FIG. 20 shows the timing of the control signal PAI in the fifth embodiment. Here, it is assumed that the start Y address (Yj) ends with “0110”. It is assumed that the number of columns accessed in the main memory array 2 is “3”, not “5”. PAI is set to the high level in the clock cycle centered at time t2, the start address Yj is loaded into the down counter 30, and "0110" appears on the address signal lines YA3, YA2, YA1, YA0. When the down counter 30 decreases, the address output value changes to “0101” at time t3 and to a value ending at “0100” at time t4. In the clock cycle centered at time t5, the access counter 50 sets the PAI to high level again, reloads the same address (Yj) from the address register 52 to the down counter 30, and the value "0110" is again set to the address signal line YA3. Appears at YA2, YA1, YA0.

  The fifth embodiment is not limited to the operation mode shown in FIG. The beginning of the burst may be the old data Db5 instead of the new data Da5, and the memory devices may be cascaded. In this case, as described in the first embodiment, it is preferable that the input of the new data Da5 is controlled to have the same timing as the first output of the old data Db5.

  Regardless of whether the beginning of the burst is the new data Da5 or the old data Db5, the reloading of the Y address from the address register 52 to the down counter 30 in response to the ADX / control signal causes the fifth implementation. In the embodiment, it is possible to receive and store the pixels of the entire horizontal scanning line by inputting a single column address, and obtain a burst of output data necessary for filtering. This is an advantage of the apparatus for controlling the fifth embodiment. This is because the controller need not repeatedly supply the same column address.

Sixth Embodiment Referring to FIG. 21, the sixth embodiment is similar to the fifth embodiment. However, the configuration of the data bus is different. The same reference numerals as those in the fifth embodiment denote the same or corresponding members. The difference from the fifth embodiment will be described below.

  In the fifth embodiment, the main data bus and the sub data bus are interconnected by the data bus switch 26. In the sixth embodiment, the main data bus and the sub data bus are not connected to each other. As in the second embodiment, the main data bus lines Dm and Dm / and the sub data bus lines Ds and Ds / are separately connected to the output amplifier 36 in the data output unit 16. Therefore, as in the second embodiment, the output amplifier 36 plays the role of the data bus switch 26 in terms of transferring data from the main data bus to the sub data bus and is written to the sub memory array 4. The write amplifier 28 plays a role in amplifying data. For this reason, the data bus switch 26 and the write amplifier 28 can be omitted.

  As in the second embodiment, the main data bus lines Dm and Dm / are connected to a latch 34 in the data input unit 14. The sub data bus lines Ds and Ds / are not connected to the data input unit 14.

  The sixth embodiment operates in the same manner as the fifth embodiment. Since there is no data bus switch 26, the electrical resistance of the transistors in the data bus switch 26 is eliminated, and access to the sub memory array 4 becomes faster. Further, as in the second embodiment, the electric capacitance of the data bus line pair can be reduced. Therefore, the sixth embodiment can operate at a higher clock rate than the fifth embodiment, and the circuit configuration is simplified by removing the data bus switch 26 and the write amplifier 28.

Seventh Embodiment Referring to FIG. 22, the seventh embodiment is similar to the fifth embodiment. However, the configuration of the column address bus is different. The same reference numerals as those in the fifth embodiment denote the same or corresponding members. Hereinafter, differences from the fifth embodiment will be described.

  In the fifth embodiment, the lower address bus 20 is directly coupled to the main Y decoder 8 and coupled to the secondary Y decoder 10 via the address bus switch 26. In the seventh embodiment, the lower address bus is doubled, one set of lower address lines 47 is coupled to the main Y decoder 8, and the other set of lower address lines 48 is connected to the sub Y decoder 10. Are combined. The address bus switch 22 and the address holding latch 40 of the fifth embodiment are omitted.

  The seventh embodiment operates in the same manner as the fifth embodiment. However, since the address bus switch 22 is not provided, the electric resistance of the transistors in the address bus switch 22 is eliminated, and access to the sub memory array 4 becomes faster. Furthermore, since the electric capacitance of the lower address bus line can be reduced, this can also speed up access. Therefore, the seventh embodiment can operate at a faster clock rate than the fifth embodiment, and the circuit configuration is simplified by removing the address bus switch 22 and the address holding latch 40.

Eighth Embodiment Referring to FIG. 23, the eighth embodiment is a combination of the features of the sixth embodiment and the features of the seventh embodiment. The main data bus lines Dm and Dm / and the sub data bus lines Ds and Ds / are separately connected to the output amplifier 36 in the data output unit 16, and separate lower address bus lines 47 and 48 are connected to the main Y decoder 8 and the sub data bus lines Ds /. It is provided for the Y decoder 10. The address bus switch 22, the data bus switch 26, the write amplifier 28, and the address holding latch 40 of the fifth embodiment are all omitted.

  The eighth embodiment has the advantages of the sixth embodiment and the seventh embodiment in terms of the simplification of the circuit configuration and the speeding up of the operation.

Ninth Embodiment Referring to FIG. 24, the memory device of the ninth embodiment has two banks A and B. Each of the two banks has the same configuration as that of the fifth embodiment. Members that are the same as or correspond to those in the fifth embodiment are indicated by the same reference numerals with suffixes A and B. Thus, bank A includes main memory array 2A, sub memory array 4A, X decoder 6A, main Y decoder 8A, sub Y decoder 10A, Y address generator 12A, upper address bus 18A, lower address bus 20A, address bus. It has a switch 22A, an internal data bus 24A, a data bus switch 26A, a write amplifier 28A, and an address holding latch 40A. On the other hand, the bank B includes a main memory array 2B, a sub memory array 4B, an X decoder 6B, a main Y decoder 8B, a sub Y decoder 10B, a Y address generator 12B, an upper address bus 18B, a lower address bus 20B, and an address bus switch 22B. And an internal data bus 24B, a data bus switch 26B, a write amplifier 28B, and an address holding latch 40B. The two data buses 24A and 24B are coupled to the data input unit 14 and the data output unit 16 via the bank switch 56, and the data input unit 14 and the data output unit 16 are shared by the banks A and B.

  The details of the circuit configuration of each bank are the same as those shown in FIG.

  FIG. 24 also shows one operation mode of the ninth embodiment. The main memory arrays in banks A and B store pixel data for one frame (consisting of two consecutive fields) of the moving image. The field data is divided into two banks as described later by even / odd distinction. The two sub memory arrays hold data delayed by one frame (2 fields) with respect to the data in the main memory array of the same bank.

  In this mode of operation, the output burst is divided into three parts.

  First, data is read from main memory array 2A in bank A, for example (operation 1). At this time, the data of one pixel is transferred to the sub memory array 4A in the bank A. Data transferred to the sub memory array 4A is two fields old. Other data read from the main memory array 2A is for the current field.

  Next, the pixel data one field older is read from the main memory array 2B in the bank B (operation 2a). At the same time, the data of one new pixel in the current field is written in the position in the main memory array 2A in the bank A which is the transfer source of the data transfer to the sub memory array 4A in the operation 1 (operation 2b). .

  Finally, pixel data is read from the sub memory array 4A in the bank A (operation 3). These data are two fields old.

  The first pixel data in the burst may be an old pixel transferred to the sub memory array 4A or a new pixel written to the main memory array 2A. This depends on whether the memory device is used in a slave connection.

  FIG. 25 shows a case where five pixels are output in each part of the burst. First, five data Da5 to Da1 belonging to the newest field are output from the main memory array 2A in the bank A. (The data Da5 is actually output by direct transfer from the data input unit 14 to the data output unit 16.) Next, the data Db5 to Db1 belonging to the previous field are stored in the main memory array in the bank B. 2B. Finally, as shown in FIG. 25 in which the data Dc5 to Dc1 belonging to the previous two fields are output from the sub memory array 4A in the bank A, the ninth embodiment uses a single memory device to Plays the role of three field memories (F1 and F2) and three sets of line memories (L11 to L34), making it possible to provide data from three consecutive fields (fields a, b, c) .

  Referring back to FIG. 2, during digital filtering, the processing operations are usually performed on pixels in an odd number of fields, including the field in which the pixel value to be filtered is generated, and before and The same number of fields after. In the ninth embodiment, the filtering operation for generating the filtered pixels in the field Fm is performed by a single memory device by processing the data in the three fields Fm, Fm-1, and Fm + 1. Realize. Here, the fields Fm, Fm-1, and Fm + 1 correspond to the fields a, b, and c, respectively.

  Due to the subordinate connection, the ninth embodiment further provides a filtering operation for generating filtered pixels in the field Fm by performing operations on data in the five fields Fm, Fm-1, Fm-2, Fm + 1, and Fm + 2. This is realized with one memory device. The first memory device receives the new pixel data Da5 of FIG. 25, and at the same time outputs the old pixel data Dc5 and a burst of pixel data from the other fields a, b and c (pixel data Dc5 is The second memory device receives the pixel data Dc5 from the first memory device and outputs the pixel data from the fields c, d, and e. Fields a, b, c, d, and e correspond to fields Fm + 2, Fm + 1, Fm, Fm-1, and Fm-2, respectively. In this case, two memory devices can be used in place of the conventional four field memories and five sets of line memories.

  26 to 29 show a method of storing field data according to the ninth embodiment.

  In FIG. 26, bank A stores pixel data of odd-numbered X addresses (row addresses) in field a (latest field) and data of pixels of even-numbered X addresses in field b (previous field). Bank B stores even X address pixel data in field a and odd X address pixel data in field b. The secondary memory array holds corresponding data in fields c and d. The numbers “1”, “2”, and “3” attached to the arrows in the drawing are operations for reading data from three pixels in a row of an odd-numbered X address in the field a (1), and the correspondence in the same row in the field b. An operation (2) of reading data of three pixels to be performed and an operation (3) of reading data of corresponding three pixels in the same row in the field (c) are shown. These operations are performed when a new pixel having an odd X address is received at a position indicated by a black dot.

  The new pixel to be received next has an even row (X) address. This pixel is stored in a location indicated by a black dot in FIG. On the other hand, the data of the three pixels having the even X address are read from field a (4), then from field b to (5), and then from field c to (6).

  With these operations, new pixels are alternately stored in bank A and bank B. These operations continue until all the pixels in field a are received.

  Referring to FIG. 28, in the next field (field z), pixel data of even X address is stored in bank A, and pixel data of odd X address is stored in bank B. In these cases, the data in field b is overwritten and the data in field a is preserved. A black dot indicates a position in bank A where data of a new pixel having an even X address is stored. The numbers “1”, “2” and “3” read the data of the three pixels of this even X address in field z (1), read the data of the corresponding three pixels in the same row of field a Operation (2) and operation (3) for reading the data of the corresponding three pixels in the same row of field b are shown.

  In FIG. 29, in the bank B, a black dot indicates a position where data of the next new pixel (having an odd X address) is stored. The numbers “4”, “5”, and “6” read the data of the three pixels at this odd-numbered X address in the field z (4), the data of the corresponding three pixels in the same row of the field a. An operation (5) for reading out and an operation (6) for reading out data of corresponding three pixels in the same row of the field b are shown.

  In FIG. 26 to FIG. 29, the even and odd rows are shown separately. This is for convenience of explanation, and of course, even and odd rows may be interleaved.

  Next, several operation modes of the ninth embodiment will be described. In the first two modes, the method outlined with reference to FIGS. 26 to 29 is used, and pixel data is output from three consecutive fields in a single burst as shown in FIG.

  FIG. 31 shows an operation mode suitable for a cascaded memory configuration.

  At time t1, CS / and RAS / are at a low level, the odd X address (Xi) is latched and decoded, and the corresponding word line (WLi) in bank A is driven by the X decoder 6A. The bank bus switch 56 is in a state of connecting the bank A to the output unit 16.

  At time t2, CS / and CAS / are low, and the Y address (Yj) is received and stored in the address register 52 in the Y address generators 12A and 12B of banks A and B. In the bank A, the address Yj is loaded into the down counter 30, the main and sub Y decoders 8A and 10A are sequentially enabled, the address bus switch 22A is controlled (similar to the first embodiment), and the old data Dcj is stored. The data is transferred from the main memory array 2A to the sub memory array 4A and to the data output unit 16. In bank B, at this time, precharging of the word lines and initialization (precharging and equalization) of the bit lines and the data bus lines are started.

  At time t3, CS / and WE / are at the low level, and new data Daj of the row address Xi and the column address Yj is received and latched in the data input unit 14. At the same time, the data output unit 16 outputs old data Dcj having the same row-column address. During this time, the down counter 30 in the Y address generator 12A continues to decrease, the data Daj-1 and Daj-2 are transferred from the main memory array 2A in the bank A to the data output unit 16, and the word line WLi is in the bank B. Is activated.

  When the transfer of the data Daj-2 from the main memory array 2A to the data output unit 16 is completed, the bank bus switch 56 is switched to connect the bank B to the output unit 16 and connect the bank A to the input unit 14. Then, the transfer of data Dbj, Dbj-1, Dbj-2 from the main memory array 2 in the bank B to the data output unit 16 starts. At this time, the down counter 30 in the Y address generator 12b generates a necessary column address.

  The burst output of data continues as shown from time t3 to t4. Output data is first read from bank A (field a) and then from bank B (field b). While data is being read from main memory array 2B in bank B, address Yj is reloaded into down counter 30 in Y address generator 12A and new data Daj is written into main memory array 2A in bank A. . When the data transfer from the bank B to the data output unit 16 is completed, the bank B is precharged again and initialized. During this time, data transfer from the sub memory array 4A in the bank A to the data output unit 16 is also started. At this time, the bank bus switch 56 couples the bank A to the output unit 16, and the down counter 30 in the Y address generator 12A again counts down from Yj to Yj-2.

  At time t4, the next X address (Xi + 1) is input. This is an even-numbered row address, and the corresponding word line (WLi + 1) in the bank B is activated. At the same time, the data output unit 16 outputs data Dcj, and the data Dcj-2 is transferred from the sub memory array 4A in the bank A to the data output unit 16.

  At time t5, data Dcj-2 is output from the data output unit 16, and the first burst ends. At the same time, CS /, CAS / and ADX / are made low, and the column address Yj is reloaded from the address register 52 into the down counter 30 in the Y address generators 12A and 12B of both banks. At this time, initialization (precharge and equalization) of the word lines, bit lines, and data bus lines in the bank A starts.

  At time t6, the second burst starts, and data is output from main memory array 2B in bank B and new data is written in main memory array 2B in bank B. The second burst is performed in the same manner as the first burst. However, the roles of bank A and bank B are switched.

  At time t7, another new X address (Xi + 2) is received, and after the end of the second burst, a third burst is performed. The third burst is performed in the same manner as the first burst by reloading the same column address using the ADX signal.

  As shown in FIG. 31, by alternately accessing banks A and B, precharging of one bank is performed while reading data from another bank, and the delay between bursts can be shortened. Writing of new input data to each bank is also performed during reading of data from other banks. This increases the efficiency of operation.

  Simultaneous input and output of data having the same X and Y addresses can be used when both memory devices are connected in cascade, and both receive the same address signal and control signal (CS /, RAS /, CAS /, WE, and ADX /). This means that output data can be obtained at the same timing from both. As a result, the design of the memory system connected in cascade is simplified.

  All X addresses in FIG. 31 are generated externally. However, the ninth embodiment can operate in the mode described in the fifth embodiment. In this case, a new X address is automatically generated in response to CS /, RAS /, and ADX /.

  FIG. 32 shows a similar series of bursts suitable for use in a memory configuration that is not cascaded. The operation in this mode is substantially the same as in FIG. However, new data (Daj) is input two clock cycles earlier, thereby allowing new data to be transferred from the data input unit 14 to the data output unit 16 and output as the first data of the burst. . In this mode, the ninth embodiment outputs data of three pixels from each of the three fields without overlapping outputs. Detailed description is omitted.

The operation mode described above in the ninth embodiment is useful in digital filtering that requires field delay. However, the ninth embodiment is also advantageous in other operations, such as motion estimation (in which case frame delay is required). Referring to FIG. 33, such an operation requires a burst of data from fields a and c and no data from field b in between.
FIG. 34 shows the output of this type of burst in subordinate mode. In the first burst started by inputting the X address at time t1, data from fields a and c are output. Data Dcj is output twice. Once, it is simultaneously with the input of new data Daj. Only one bank, eg bank A, is used in this burst. The other bank (Bank B) is precharged between times t3 and t4 during the first burst. Command and data input for the second burst using bank B occurs at times t4, t5, t6. This initiates the second burst immediately (without interruption) after the end of the first burst. The address for the second burst is generated by the ADX / signal as described in the fifth embodiment. During the second burst, new data Daj is written to the main memory array 2A in the bank A as a background operation, and the bank A is precharged.

  FIG. 35 shows a similar non-dependent connection operation mode of the ninth embodiment. The operation timing is substantially the same as in FIG. However, new data (Daj) is input two clock cycles earlier, allowing new input data to be transferred from the data input unit 14 to the data output unit 16 and output as the first data of the burst. . Each burst consists of 5 pixels from each of 2 fields (with 1 frame delay between each other).

  Some types of image processing operations require different amounts of data for different fields. FIG. 36 shows a case where 5 pixels from field a, 3 pixels from field B, and 5 pixels from field c are required. Since the ninth embodiment has separate Y address generators 12A and 12B for banks A and B, the above request can be obtained by separately controlling the down counters and Y address decoders of banks A and B. Can be easily met.

  FIG. 37 shows this type of burst output in non-dependent connection mode. This burst includes 5 pixels from field a (Daj to Daj-4), 3 pixels from field b (Dbj to Dbj-2), and 5 pixels from field c (Dcj to Dcj-4). This type of burst is repeated with a period of 14 clock cycles. This is equal to the spacing between the new pixel inputs (indicated by the letter “T”).

  Further details of this type of operation will appear later in the description of the embodiment.

Tenth Embodiment Referring to FIG. 38, the tenth embodiment is similar to the ninth embodiment. However, as in the sixth embodiment, each bank has separate data bus lines for the main and sub memory arrays. Bank A has a main data bus 58A and a sub data bus 60A. Bank B has a main data bus 58B and a sub data bus 60B. The main data buses 58A and 58B are connected to the data input unit 14 and the data output unit 16 via the bank bus switch 56. The sub data buses 60A and 60B are connected to the data output unit 16 via the bank bus switch 56.

  The other members in FIG. 38 are the same as those in FIG. 24, and are denoted by the same reference numerals.

  Since separate main and sub data buses are provided, the data bus switches 26A and 26B and the write amplifiers 28A and 28B of the ninth embodiment are not required, and the circuit configuration is simplified. Furthermore, as in the sixth embodiment, the electrical resistance and capacitance of the data bus line are reduced, and the operation speed is increased.

Eleventh Embodiment Referring to FIG. 39, the eleventh embodiment is similar to the ninth embodiment. However, as in the seventh embodiment, each bank has separate address bus lines for the main and sub memory arrays. Bank A has a main lower address line 47A and a sub-lower address line 48A. Bank B has a main lower address line 47B and a sub lower address line 48B. Other members in FIG. 39 are the same as those in FIG. 24, and are denoted by the same reference numerals.

  By providing separate main and sub address buses, the address bus switch and the address holding latch of the ninth embodiment are not required, and the circuit configuration is simplified. Furthermore, as described in the seventh embodiment, the electrical resistance and capacitance of the address bus line can be reduced, and the operation speed is increased.

Twelfth Embodiment Referring to FIG. 40, the twelfth embodiment is a combination of the features of the tenth and eleventh embodiments. Separate main data buses 58A, 58B and separate sub data buses 60A, 60B are provided, and separate main lower address lines 47A, 47B and separate sub lower address lines 48A, 48B are provided. The bus switches 22A, 22B, 26A, and 26B, the write amplifiers 28A and 28B, and the address holding latches 40A and 40B of the ninth embodiment are all unnecessary.

  The twelfth embodiment has the advantages of both the tenth and eleventh embodiments (simplified circuit configuration and faster operation).

Thirteenth Embodiment Referring to FIG. 41, the thirteenth embodiment is similar to the eleventh embodiment. However, the data buses 24A and 24B are not divided into main and sub parts, and no write amplifier is provided. Thus, the bus switches 26A and 26B and the write amplifiers 28A and 28B of the eleventh embodiment are eliminated.

  As a new member, the thirteenth embodiment includes a transfer register 62 coupled to the bank bus switch 56. The transfer register 62 temporarily stores data being transferred from the main memory array in one bank to the sub memory array, so that writing of the data to the sub memory array is performed by other banks being accessed. It is possible to perform as a background motion while

  Other members are the same as those in FIG. 39, and are denoted by the same reference numerals. The interconnection of data input unit 14 and data output unit 16 is clearly shown in FIG.

  Next, the operation of the thirteenth embodiment in the non-dependent connection and field delay mode will be described with reference to FIGS. Since the timing relationship between the input and output data and the control signal is the same as the corresponding mode of the ninth embodiment, reference is also made to the timing and data values shown in FIG.

  As shown in FIG. 32, after the inputs (t1, t2) of the X and Y addresses Xi and Yj, the word lines WLai and WLbi in the banks A and B in FIG. 42 are at the timing described in the ninth embodiment. The bank bus switch 56 couples the data output unit 16 and the transfer register 62 together to the bank A data bus 24A. The old data Dcj stored in the XY address (Xi-Yj) in the main memory array 2A of the bank A is transferred to the transfer register 62 via the data bus 24A and the bank bus switch 56 (arrows in FIG. 42). ), The new input data Daj is transferred from the data input unit 14 to the data output unit 16. Data Daj is output from the data output unit 16 with a read latency of 2 clock cycles (FIG. 32, time t3).

  Referring to FIG. 43, after data Dcj, data Daj-1 and Daj-2 are transferred from main memory array 2A to data output unit 16 via data bus 24A and bank bus switch 56, and the next after time t3. It is output from the data output unit 16 in two clock cycles. The data Daj and Dcj are held in the data input unit 14 and the transfer register 62, respectively.

  Referring to FIG. 44, following the transfer of data Daj-2 from main memory array 2A in bank A to data output unit 16, bank bus switch 56 is switched, and data input unit 14 is transferred to data bus 24A in bank A. The data output unit 16 is connected to the data bus 24B of the bank B. Both Y address generators 12A and 12B generate a column address Yj. New input data Daj is transferred from the data input unit 14 to the main memory array 2A in the bank A via the bank bus switch 56 and the data bus 24A, and is written in the memory cell in which the data Dcj is stored. At substantially the same time, the data Dbj is transferred from the main memory array 2B in the bank B to the data output unit 16 via the data bus 24B and the bank bus switch 56, and is output next to the data Daj-2.

  Next, referring to FIG. 45, the Y address generator 12A in the bank A generates an arbitrary convenient column address. On the other hand, the Y address generator 12B in the bank B counts down and outputs Yj-1 and Yj-2. In these clock cycles, the data Dcj held in the transfer register 62 is transferred to the sub memory array 4A in the bank A via the bank bus switch 56 and the data bus 24A, and is designated by the Y address generator 12A. Written in the memory cell. On the other hand, the data Dbj-1 and Dbj-2 are transferred from the main memory array 2B in the bank B to the data output unit 16 via the data bus 24B and the bank bus switch 56. These data are output from the data output unit 16 next to the data Dbj.

  Referring to FIG. 46, the data to be transferred to the data output unit 16 and to be output from the data output unit 16 is data Dcj held in the transfer register 62. Data Dcj is output from data output unit 16 at time t4 in FIG.

  Referring to FIG. 47, bank bus switch 56 is switched again, data output unit 16 is connected to data bus 24A in bank A, and Y address generator 12A further generates two column addresses. The data Dcj-1 and Dcj-2 are transferred from the sub memory array 4A in the bank A to the data output unit 16 via the data bus 24A and the bank bus switch 56, and output following the data Dcj. This completes the first burst of FIG.

  The next burst is performed in the same manner. However, the roles of banks A and B are switched.

  The thirteenth embodiment can also operate in the slave connection mode as shown in FIG. The internal operation is the same as that described with reference to FIGS. However, transfer of new data Daj from the data input unit 14 to the data output unit 16 is not performed. Instead, the old data Dcj is transferred to the data output unit 16 and output at the beginning of the burst, and the new data Daj is input simultaneously with the output of the old data Dcj (FIG. 48).

The sequence and timing of external data input and output in the thirteenth embodiment are the same as in the ninth embodiment, but the transfer register 62 relaxes the demand for internal operations in several ways. As shown in FIG. 45, the writing of data Dcj from the transfer register 62 to the sub memory array 4A is performed in the background during the transfer of data Dbj-1 and Dbj-2 from the main memory array 2B to the data output unit 16. As a result, the timing margin for amplification of the data Dcj on the bit line in the sub-M memory array 4A is increased. Further, since transfer of data Dcj from main 2A and transfer to sub memory array 4A are performed in separate operations, when access to sub memory array 4A starts, down counter 30 in Y address generator 12A There is no need to reload the down counter 30, and the down counter 30 may continue counting down following the current value. The down counter 30 in the Y address generator 12A performs a reload once (not twice as in the ninth embodiment) before writing the data Daj into the main memory array 2A during the entire burst. Need to do.

  In the above description, the data Daj is written into the main memory array 2A before the data Dcj is written into the sub memory array 4A. However, these orders may be reversed.

  As a modification of the thirteenth embodiment, data can be stored at an arbitrary address in the sub memory arrays 4A and 4B, and the data buses 24A and 24B are not divided into main and sub parts. The embodiment can be applied to the case where there is only one memory array, Y decoder, and Y address bank in each bank. In this case, a part of the memory array of each bank is used as a sub-part, and the overwritten data is transferred from the other (main) part of the same array to this sub-part.

Next, five further embodiments related to the thirteenth embodiment will be described. In the following description, the same reference numerals as those in FIG. 41 denote equivalent members.

Fourteenth Embodiment Referring to FIG. 49, the fourteenth embodiment includes an input data register 64 coupled between the data input unit 14 and the bank bus switch 56. Input data is transferred from the data input unit 14 to the input data register 64 and held in the input data register 64 until it is written to one of the main memory arrays.

  Except for this difference, the fourteenth embodiment operates in the same manner as the thirteenth embodiment. Therefore, detailed description is omitted.

  The input data register 64 increases the degree of freedom regarding the timing of the internal operation of the memory device. Another advantage is that the input data is stored near the memory array. This is important because there may be a long distance between the data input unit 14 and the bank bus switch 56 in terms of signal line length and signal propagation time.

Fifteenth Embodiment Referring to FIG. 50, the fifteenth embodiment is the same as the thirteenth embodiment. However, instead of the single transfer register 62, separate transfer registers 62A and 62B are coupled to the data buses 24A and 24B of the banks A and B. With this configuration, data can be transferred from the main memory array of each bank to the transfer register and from the transfer register to the sub memory array of each bank without going through the bank bus switch 56. Therefore, the electric resistance of the bank bus switch 56 does not become an obstacle during the transfer.

Sixteenth Embodiment Referring to FIG. 51, the sixteenth embodiment is a combination of the features of the fourteenth embodiment and the fifteenth embodiment. That is, it has an input data register 64 as in the fourteenth embodiment and separate transfer registers 62A and 62B as in the fifteenth embodiment. The advantages of this configuration are as described above.

Seventeenth Embodiment Referring to FIG. 52, in the seventeenth embodiment, a single full-duplex data bus 66 is used for both banks A and B, thereby the thirteenth embodiment. The bank bus switch 56 is removed. The full-duplex data bus is a data bus that can perform reading and writing at the same time without causing a collision. Data bus 66 is directly coupled to data input unit 14, data output unit 16 and transfer register 62.

  Other members are the same as those in the thirteenth embodiment, and are denoted by the same reference numerals.

  The seventeenth embodiment operates in substantially the same manner as the thirteenth embodiment. However, there is no need to switch the data bus, and there is no electrical resistance of the bank bus switch.

Eighteenth Embodiment Referring to FIG. 53, the eighteenth embodiment is a combination of the features of the fourteenth embodiment and the seventeenth embodiment. That is, it has a full-duplex data bus 66 coupled to the data output unit 16, transfer register 62, and input data register 64. The advantages of this configuration are as described above.

Nineteenth Embodiment Referring to FIG. 54, the nineteenth embodiment differs from the thirteenth embodiment in the configuration of Y address generators 12A and 12B. In the thirteenth embodiment, each Y address generator has an access counter 50 separately, thereby controlling the number of addresses generated by the down counter 30. In the nineteenth embodiment, the Y address generators 12A and 12B share the same access counter 50. The shared access counter 50 is controlled by a memory control signal generator (SG) 68 (which has been omitted in the above embodiments for the sake of simplicity). In the nineteenth embodiment, an access count register 70 is also provided. The access count register 70 stores an initial value ADN for the access counter 50. This value is supplied from the memory control signal generator 68.

  Other members in the nineteenth embodiment (including the transfer register 62) are the same as those in the thirteenth embodiment, and are denoted by the same reference numerals.

  The nineteenth embodiment provides an efficient method for controlling the address generator such that each of the Y address generators 12A, 12B always generates a fixed number of consecutive addresses. The memory control signal generator 68 simply writes the appropriate fixed number ADN to the access count register 70 and then commands the access counter 50 to reload this fixed number ADN at the appropriate time, Control is performed on the address generators 12A and 12B.

  This control mode is useful, for example, in filtering operations that require the same number of pixels from each of the three fields, as shown in FIGS.

Twenty Embodiment Referring to FIG. 55, the twentieth embodiment is a combination of the features of the fourteenth embodiment and the nineteenth embodiment, and includes a transfer register 62, an input data register 64, A shared access counter 50 controlled by a memory control signal generator 68 and an access count register 70 for storing a fixed value ADN are provided. The fixed value ADN stored in the access count register 70 is used by the access counter 50 to control the number of consecutive addresses generated by the Y address generators 12A and 12B. The advantages of the twentieth embodiment are as described above.

Twenty-first embodiment Referring to FIG. 56, the twenty-first embodiment is a combination of the features of the fifteenth embodiment and the features of the nineteenth embodiment. Instead of a single transfer register 62 of the form, it has separate transfer registers 62A, 62B coupled to the data buses 24A, 24B in banks A, B. The advantages of the twenty-first embodiment are as described above.

Twenty-second Embodiment Referring to FIG. 57, the twenty-second embodiment is a combination of the features of the sixteenth and nineteenth embodiments, and includes an input data register 64 and a separate bank A. , B, separate transfer registers 62A, 62B coupled to the data buses 24A, 24B, and other members described in the nineteenth embodiment. The advantages of the twenty-second embodiment are as described above.

Twenty-third Embodiment Referring to FIG. 58, the twenty-third embodiment is a combination of the features of the seventeenth and nineteenth embodiments, and includes a data input unit 14 and a data output unit 16. And a full-duplex data bus 66 coupled to the transfer register 62 and the other members described in the nineteenth embodiment. The advantages of the twenty-third embodiment are as described above.

Twenty-fourth Embodiment Referring to FIG. 59, the twenty-fourth embodiment is a combination of the features of the eighteenth embodiment and the nineteenth embodiment, and includes a data output unit 16, a transfer register 62, And a full-duplex data bus 66 coupled to the input data register 64 and the other members described in the nineteenth embodiment. The advantages of the twenty-fourth embodiment are the same as described above.

Twenty-fifth Embodiment Referring to FIG. 60, the twenty-fifth embodiment is obtained by adding two access count registers 72 and 74 to the configuration of the nineteenth embodiment. The first access count register 70 holds a value ADN1 corresponding to the number of pixels required from the first accessed field in the burst. The second and third access count registers 72 and 74 hold values ADN2 and ADN3 corresponding to the number of pixels required from the second and third fields. All three values ADN1, ADN2, and ADN3 are supplied from the memory control signal generator 68.

  As shown in FIGS. 36 and 37, the twenty-fifth embodiment is useful in a filtering operation in which the number of required pixel data is different for each field. For the operations of FIGS. 36 and 37, ADN1, ADN2, and ADN3 are set to “5”, “3”, and “5”, respectively. ADN1 is used to control the first part of the burst (at which time the data is read from the main memory array 2A of bank A, for example). ADN2 is used to control the second part of the burst (at which time the data is read from, for example, main memory array 2B in bank B). ADN3 is used to control the third part of the burst (at which time the data is read from, for example, the sub-memory array 4A of bank A).

Twenty-sixth Embodiment Referring to FIG. 61, the twenty-sixth embodiment is obtained by adding a pair of address recalculators (RECALC) 76A and 76B to the configuration of the twenty-fifth embodiment. The address recalculator 76A is coupled to the Y address generator 12A in the bank A, and determines the start Y address used by the Y address generator 12A in response to the shift control signal SFTa output from the memory control signal generator 68. Recalculate. The address recalculator 76B is coupled to the Y address generator 12B in the bank B, and determines the start Y address used by the Y address generator 12B in response to the shift control signal SFTb output from the memory control signal generator 68. Recalculate.

  The twenty-sixth embodiment is particularly useful in filtering used for conversion from interlaced scanning to sequential scanning. This is illustrated in FIGS. 62-64. 62-64 show vertical cross sections of several successive fields distributed along the time axis.

  In FIG. 62, in interlaced scanning, pixels in even fields (fields a and c) are arranged in a space between pixels in odd fields (fields b and d). In a typical filtering process to generate a filtered pixel value at the location of pixel Dc3, five pixels from even field c (Dc1 to Dc5) and four pixels from previous odd field d (Dd2 Dd5) and four pixels (Db2 to Db5) from the later odd field b are required.

  These data are generated as a single burst in the twenty-fifth embodiment. This is because when stored in the memory device, the pixels Db5, Dc5, Dd5 all have the same column address as indicated by the arrows. That is, the start column address is the same in these three fields b, c, and d. All that is required is to give the value “4” to ADN1 and ADN3 and the value “5” to ADN2. The symbol “x” in pixel Db5 indicates where new data is rewritten onto old data during a burst.

  For conversion to progressive scanning, generation of a new pixel located in the middle of the pixels Dc2 and Dc3 in the field c is required, and the data shown in FIG. 63, that is, the pixels Dd1 to Dd5 in the field d and the pixels in the field c Dc1 to Dc4 and pixels Db1 to Db5 in the field b are required. It is not easy to generate these data in the twenty-fifth embodiment. This is because the start address (column address Dc4) of field c is different from the start addresses (Dd5, Db5) of fields b and d.

  FIG. 64 shows pixel data necessary for conversion of the pixel value Db3 in the odd field b, that is, Dc1 to Dc4 of the field c, Db1 to Db5 of the field b, and Da1 to Da4 of the field a. In this case, not only the start column address of the field b is different from the start column addresses of the fields a and c, but also the new input pixel data Da5 does not form part of the output burst.

  The twenty-sixth embodiment can deal with all these cases. The operation of FIGS. 62 to 64 will be described in detail below.

  For the burst of FIG. 62, the memory control signal generator 68 sets ADN1 to “4”, ADN2 to “5”, ADN3 to “4”, and sets both SFTa and SFTb to “0”. To do. FIG. 65 shows the resulting output sequence.

  Assuming that the output starts from bank B, ADN 1 (“4”) is first read from access count register 70 and written to access counter 50. Therefore, the access counter 50 causes the Y address generator 12B in the bank B to generate four Y addresses (Yj to Yj-3). Data Dd5 is transferred from main memory array 2B in bank B to transfer register 62, new input data Db5 is transferred from data input unit 14 to data output unit 16, and data Db4 to Db2 are transferred to main memory array 2B in bank B. Is output from the data output unit 16 after the data Db5.

  Next, ADN 2 (“5”) is read from second access count register 72 and written to access counter 50. Therefore, the access counter 50 causes the Y address generator 12A in the bank A to generate five Y addresses (Yj to Yj-4). Dc5 to Dc1 are sequentially output from the main memory array 2A in the bank A. Meanwhile, in the background, the data Dd5 is written from the transfer register 62 to the sub memory array 4B of the bank B, and new input data Db5 held in the data input unit 14 is written into the main memory array 2B.

  Finally, ADN 3 (“4”) is read from the third access count register 74 and written to the access counter 50. The Y address generator 12B further generates four Y addresses, and the data Dd5 to Dd2 are output from the sub memory array 4B in the bank B.

  For the burst of FIG. 64, assume that the output begins in bank A. The memory control signal generator 68 sets ADN1 to “4”, ADN2 to “5”, ADN3 to “4”, SFTa to “1”, and SFTb to “0”. FIG. 66 shows the resulting output sequence.

  First, ADN 1 (“4”) is read from the access count register 70 and written to the access counter 50. Therefore, the Y address generator 12A generates four Y addresses. Since SFTa is “1”, the address recalculator 76A calculates the start address, and determines Yj−1 instead of Yj. The Y address generator 12A skips the address Yj and starts outputting the Y address from Yj-1. Data Da4 to Da1 are transferred from the main memory array 2A in the bank A to the data output unit 16 and output by the data output unit 16 as the first part of the burst.

  Next, ADN 2 (“5”) is read from second access count register 72 and written to access counter 50. Therefore, the access counter 50 causes the Y address generator 12B in the bank A to generate five Y addresses. Since SFTb is “0”, the address recalculator 76B does not change the start address, and therefore the start address remains Yj. Db5 to Db1 are sequentially output from the main memory array 2B in the bank B and output by the data output unit 16. Meanwhile, in the bank A, in the background, the Y address generator 12A generates the previously skipped address Yj. Data Dc5 is transferred from main memory array 2A to transfer register 62, and the new input data Da5 is written to the same address in main memory array 2A. After this write operation, the Y address generator 12A generates an appropriate address, and the data Dc5 held in the transfer register 62 is written into the sub memory array 4A in the bank A.

  Finally, ADN3 (“4”) is written to the access counter 50, the start address in the Y address generator 12A is recalculated by the address recalculator 76A, and the Y address generator 12A is obtained by recalculation. Four Y addresses starting from the start address are generated, and the data Dc4 to Dc1 are output from the sub memory array 4B in the bank B.

  For the burst of FIG. 63, when the burst starts from bank B, the memory control signal generator 68 sets ADN1 to “5”, ADN2 to “4”, ADN3 to “5”, and SFTa to “ 1 and SFb are set to “0”, respectively. Detailed description is omitted.

  The above operation using a different number of pixels for each field is useful not only for conversion from interlace scanning to sequential scanning, but also for picture-in-picture processing, format conversion, noise removal, and the like.

  Next, modifications that apply to all of the above embodiments will be described.

  Referring to FIG. 67, in this modification, the main memory array 2 and the sub memory array 4 have separate X decoders and word lines. The sub memory array 4 has its own X decoder 77. If there are two banks, the sub-memory array of each bank has a separate X decoder.

  The X and Y decoders are generated by an XY address generator 78. The XY address generator supplies the X address to the X decoders 6 and 77 via the X address bus 79, and the Y address is an upper Y address. This is supplied to the bus 18 and the lower address bus 20.

  The internal data bus 24 is divided into two parts by a data bus switch 80. The data bus switch 80 selectively couples the main data bus (MDB) to the data input unit 14 and the data output unit 16, and selectively couples the sub data bus (SDB) to the data output unit 16, The sub data buses are selectively coupled to each other.

  When the X address (Xi) is received, the X decoders 6 and 77 simultaneously activate the corresponding word lines WLi and WLi 'of the main and sub memory arrays. Since the two word lines WLi and WLi 'are activated simultaneously, they can be considered as different parts of the same word line. Similarly, X decoders 6 and 77 operate as if they were a single X decoder that produces a double decoded output.

  Since both the X decoders 6 and 77 receive the same X address, this modification operates in substantially the same manner as in the first embodiment, and operates in the same manner as in the other embodiments described so far. It can also be modified.

  The embodiment described so far shows a method for obtaining a data block including data from a currently processed column and data from a previously processed column, as shown in FIG. The following embodiments show a similar method for obtaining a data block containing data for a column currently being processed and data for a column to be processed later, as shown in FIG.

  In this case, there is no point in transferring the overwritten pixel data to the sub memory array. This is because an interval close to one field elapses before the overwritten data is needed again, and the data disappears during this interval. Rather, the overwritten data should be transferred to another field memory. That is, the field memories need to be cascaded as shown in FIG.

  The following embodiment exemplifies the second invention of the present application, and does not have the sub-memory array of the above-described embodiment, while the features used for the slave connection, for example, separate data input and data output It has a terminal. These embodiments provide at least the functions of the field memory F1 and the line memories L21 to L24 surrounded by a dotted line in FIG.

  The same reference numerals are given to the same or equivalent parts as those of the previous embodiment.

27th Embodiment Referring to FIG. 68, the 27th embodiment is a memory array 2, an X decoder 6, a Y decoder (YD) 8, a Y address generator 12, a data input unit 14, and a data output unit 16. And an internal data bus 24, a memory control signal generator 68, an address input unit (ADIN) 81, a data bus switch (SW) 82, and a buffer circuit (BUF) 83. The Y address generator 12 includes a down counter 30 and an access counter 50.

  As in the above embodiment, the circle Nij in the memory array 2 represents a group of memory cells that share the X address (Xi) and the Y address (Yj).

  The address input unit 81 (omitted for simplification of the drawing in the drawings of the above-described embodiment) receives the X address and the Y address separately from the external address terminal (ADD), and receives the internal address signal XAD. And YAD. The Y address signal YAD is supplied to the down counter 30, which generates a Y address YADD that is counted down from YAD and supplied to the Y decoder 8. Both the down counter 30 and the access counter 50 operate in synchronization with the internal clock signal (CLK ′). The internal clock signal (CLK ′) is generated from the external clock signal (CLK), and has the same frequency as the external clock signal in the present embodiment. The access counter 50 receives the burst length control signal PA from the address input unit 81 and generates a control signal PW for controlling the down counter 30 and the data bus switch 82.

  The data input unit 14, data bus switch 82, buffer circuit 83, and data output unit 16 are controlled by a memory control signal generator 68. The data input unit 14 transmits input data from the external data input terminal DIN to the internal data bus 24. Data bus switch 82 transmits data from data bus 24 to buffer circuit 83. This data is stored in the buffer circuit 83 until it is output from the external data output terminal DOUT by the data output unit 16. The buffer circuit 83 (omitted in the drawings of the above-described embodiment) is a FIFO buffer whose depth is equal to the longest read latency of the memory device.

  The memory control signal generator 68 has an internal mode register (not shown). It is designed so that read latency and various other access modes can be specified. The mode register is programmed with commands formed by the combination of the values received on the CS /, CAS /, RAS /, WE / signal and address input lines. This is generally similar to the mode register programming in existing memory devices such as the Oki MSM54V24632A.

  Referring to FIG. 69, address input unit 81 has three internal registers R1, R2, and R3. These are all coupled to the address input terminal ADD. When the memory control signal generator 68 receives the burst length programming command, the value received at the address input terminal is latched in the register R1. At other times, the value received at the address input terminal is latched in the register R2 if CS /, CAS / is activated, and is stored in the register R3 if CS /, RAS / is activated. Latched. The X address XAD is output from the register R3, the Y address YAD is output from the register R2, and the burst length control signal PA is output from the register R1.

  Referring to FIG. 70, access counter 50 includes a counter 84 and a control circuit 86. Counter 84 receives internal clock signal CLK 'and burst length control signal PA. The control circuit 86 receives the control signal PA and the output of the counter 84, and generates a PW control signal.

  The PA signal is a non-zero signal that causes the P control circuit 86 to activate the PW control signal, while the value of the PA signal is loaded into the counter 84. The counter 84 counts down from the PA value in synchronization with the CLK ′ signal. When the output of counter 84 reaches zero, the control circuit deactivates the PW control signal.

  Next, an operation in a preferable mode of the 27th embodiment will be described.

  When the burst length is programmed, the burst length is input from the address input terminal (ADD) and latched in the register R1 in the address input unit 81. In a personal computer or workstation, this programming can be performed by BIOS. In the embodiment described above, the burst length is indirectly limited by the size of the sub memory array. However, in the present invention, the burst can be designated to an arbitrary length. In the following description, the burst length is “5”.

  After the mode programming, an access operation is performed as shown in FIG.

  At time t1, CS / and RAS / are at a low level, and the row address (Xi) is received at the address input terminal (ADD). The row address is latched in the register R3 at the rising edge of the internal clock signal CLK '(this is substantially the same as the rising edge of the external clock signal CLK). Necessary internal latch signals are generated by the memory control signal generator 68. Memory control signal generator 68 causes X decoder 6 to decode the X address and activate the corresponding word line in memory array 2. All memory cells connected to the word line transmit data to the corresponding bit line.

  At time t2, CS /, CAS / are at a low level, and the column address (Yj) is received and latched at the address input terminal. The column address (Yj) is immediately supplied from the address input unit 81 to the down counter 30 as the column address signal YAD, and from the down counter 30 to the Y decoder 8 as the column address signal YADD. The Y decoder 8 outputs a decoded signal (shown as a waveform Yj in FIG. 71). The decoded signal activates the transfer transistor, and the complementary bit line of the column Yj is coupled to the internal data bus 24 via the transfer transistor (as described in the first embodiment). Therefore, the data Dbj held in the memory cell Nji is transmitted to the data bus 24.

  At time t2, the address input unit 81 generates a PA signal (not shown), and the access counter 50 activates the PW control signal by this PA signal. By this signal, the data bus switch 82 is closed, so that the data Dbj is transmitted to the buffer circuit 82. The buffer circuit 83 stores the data Dbj by a further control signal from a memory control signal generator 68 (not shown) generated at or before time t3.

  At time t3, the count value Yj of the down counter 30 decreases from Yj to Yj-1, and the Y decoder 8 disconnects the complementary bit line of column Yj from the data bus 24, and instead replaces the complementary bit line of column j-1 to the data bus 24. Connect to. This is shown by the change of the waveform Yj from the high level to the low level and the change of the waveform Yj-1 from the low level to the high level slightly after time t3 in FIG. The control signal PW remains at a high level. Therefore, the data bus switch 82 is kept closed, and the data Dbj-1 held in the memory cell Nij-1 is stored in the buffer circuit 83. The counter 84 in the access counter 50 decreases from “5” to “4”.

  At time t4, the count value of the down counter 30 decreases from Yj-1 to Yj-2, and the data Dbj-2 held in the memory cell Nji-2 is similarly passed through the data bus 24 and the data bus switch 82. To the buffer circuit 83. The counter 84 decreases its count value from “4” to “3”.

  Until this time, the data output unit 16 is in a high impedance state, and no data is output from the memory device. At the next rise of clock CLK at time t4, memory control signal generator 68 gives a command to data output unit 16 to start outputting data Dbj stored in buffer circuit 83 after time t2. These data Dbj appear at the data output terminal DOUT (A) at the rising edge of the external clock signal CLK at time t5, and the external device can read them at this time. At time t5, the count value of down counter 30 decreases from Yj-2 to Yj-3, data Dbj-3 held in memory cell Nij-3 is transferred to buffer circuit 83, and counter 84 has its count value. Decreases from “2” to “1”.

  Similarly, at time t6, the data output unit 16 outputs the data Dbj-1 from the buffer circuit 83, the down counter 30 decreases the count value to Yj-4, and the data Dbj-4 is transferred from the memory cell Nji-4. The count value of the counter 84 is decreased from “2” to “1”.

  At time t7, the data output unit 16 outputs data Dbj-2. Further, the count value of the counter 84 decreases from “1” to “0”, whereby the control circuit 86 in the access counter 50 inactivates the control signal PW. If the PW signal is inactive, the down counter 30 stops and the data bus switch 82 opens. Therefore, transfer of new data to the buffer circuit 83 is not performed. The Y address YADD output by the down counter 30 remains Yj-4, and the Y decoder 8 keeps the complementary bit lines of the column Yj-4 connected to the data bus 24.

  At time t8, the data output unit 16 outputs data Dbj-3. At the fall of the external clock signal CLK immediately after time t8, CS / and WE / become low level, and new input data Daj-4 to be stored in the memory array 2 is received at the data input terminal DIN (A).

  At time t9, the data output unit 16 outputs data Dbj-4, and the data input unit 14 sends new data Daj-4 to the memory array 2 via the data bus 24. At this time, since the complementary bit line of the column Yj-4 is still coupled to the data bus 24, the data Daj-4 is stored in the memory cell Nij-4 instead of the data Dbj-4 just outputted. .

  When the output terminal DOUT (A) of the memory device is coupled to the input terminal DIN (B) of another memory device having the same configuration, the second memory device receives the same write enable signal WE /, and at time t9, When the first memory device (A) stores the new data Aaj-4, the second memory device (B) receives and stores the data Dbj-4 output from the first memory device (A). . This is shown at the bottom of FIG.

  The second memory device (B) receives the data Dbj-1 to Dbj-3 output from the first memory device (A) before time t9, but at this time, the signal WE / is inactive. Therefore, these data Dbj-1 to Dbj-3 are ignored.

  After an appropriate time has elapsed from time t9, the memory control signal generator 68 gives a command to the Y decoder 8 to disconnect all bit lines from the data bus, and to the X decoder 6 all word lines are disabled. Give a command to activate. The Yj-4 signal in FIG. 71 is at a low level at this time, and the data bus 24 is initialized in preparation for the next access. In FIG. 71, this next access is another similar burst starting from the same column address (Yj) and the next row address (Xi + 1).

  In order to see the above operation in more detail, FIG. 72 shows a part of the memory array 2 used for storing one field or one frame of the moving picture P. Pixel data is stored in n rows and m columns. Here, n is the number of pixels on one horizontal scanning line, and m is the number of horizontal scanning lines in one field or frame.

  FIG. 73 shows how the contents of the memory change as the pixel data of the first horizontal scanning line is received, and shows which pixel data is read out. Scanning is done in order from left to right, and the row address or X address starts at zero and increases by one. The column address (Y address) also starts from zero as shown. The letter t represents time, P1, P2,. . . Pn indicates a state in which the memory contents change. The hatched dots indicate the pixel data already stored (just stored) in the memory array 2, the white dots indicate the old pixel data of the previous field, and the dots with the symbol “x” will be overwritten. The old pixel data is shown.

  Before being overwritten, the old data (dots marked "x") is read (indicated by a rectangular frame) along with the four old data immediately below it. For example, in the state P4, before the new data is overwritten on the old data in the column zero, the row address is “3” (Xi = 3) and the column address is “4” to “0” (Yj = 4, Yj− 4 = 0) pixel data is read out as a single burst.

  Since the column address Yj supplied to the memory device for this operation is not “0” but “4”, the input data is not stored in the column specified by the input column address. This is not disadvantageous and is even convenient in processing video signals where the first few scan lines are within the vertical blanking period and there is no data to be stored there.

  FIG. 74 similarly shows changes (Pn + 1 to P2n) when the pixel data of the second horizontal scanning line is received. In this case, the memory device receives column address “5” (Yj = 5), outputs pixel data from column “5” to column “1” as a burst, and writes new data to column “1”. Data in column “0” (Yj = 0) is not read until the beginning of the next field. As described above, this is the reason why the data of the column “0” is not stored in the sub memory array having only a storage capacity of a part of one field.

  The above description also applies to the case of progressive scanning. In this case, it is necessary to read by replacing “field” with “frame”.

  FIG. 75 shows a matrix of D-type flip-flops 88 that can be used to receive data output in the twenty-seventh embodiment. The memory A receives and outputs data as shown in FIG. Data output from the memory A is going to be transmitted to the memory B and the first flip-flop 88. The flip-flop 88 and the four flip-flops immediately below the flip-flop 88 are controlled in timing by a burst clock signal (BCLK) synchronized with the data output from the memory A. Burst clock signal BCLK is generated by applying a gate to the clock signal shown in FIG. Thus, for example, in the first burst, BCLK consists of five clock cycles from time t5 to time t9. The five flip-flops whose timing is controlled by BCLK form a shift register that stores data Db5 to Db1 during a burst.

  The other flip-flop of FIG. 75 is controlled in timing by a row clock signal (RCLK) generated from the RAS / or CAS / control signal. RCLK is pulsed once before the start of each burst. The four flip-flops to the right of the first flip-flop 88 hold the data Db11, Db21, Db31, Db41 on the left side of the data Db1 (these were output from the memory A in the four previous bursts).

  FIG. 76 shows the cascade connection of the three memory devices A, B, and C according to the twenty-seventh embodiment. Each memory device is shown as having eight data input terminals (DIN) and eight data output terminals (DOUT). Output terminal DOUT (A) of memory A is coupled to input terminal DIN (B) of memory B. The output terminal DOUT (B) of the memory B is coupled to the input terminal DIN (C) of the memory C. Similarly, any number of memory devices can be cascaded.

  Output data Db1 to Db5 shown on the right side of the memory A represent one burst output stored in the D-type flip-flop 88 whose timing is controlled by the BCLK signal shown in FIG. The D-type flip-flop whose timing is controlled by the RCLK signal shown in FIG. 75 is shown by a rectangular block having a sign of D-FF × 4 in FIG.

  Compared to the conventional SDRAM, the twenty-seventh embodiment has several advantages in this kind of subordinate connection configuration.

  One advantage is that a single XY address input is sufficient for both burst read access and write access.

  Another advantage is that data having the same X and Y addresses can be input and output simultaneously. The same address signal, write enable signal, and other control signals may be supplied to all of the memory devices that are cascade-connected at the same timing. Being able to perform input and output simultaneously means that bursts including read and write accesses are completed in a shorter time than conventional SDRAM.

  A third advantage is that the data output terminal of each memory device can be directly coupled to the data input terminal of the next memory device in the slave connection. The conventional SDRAM has only one set of data terminals, which are used for both input and output. When a conventional SDRAM is used in a slave connection configuration, it is necessary to insert a switch in between and separate input data from output data, and a separate control signal is required to control the switch.

  Because of these advantages over the conventional SDRAM in the cascade connection configuration, the twenty-seventh embodiment requires less hardware and can operate at high speed.

Twenty-eighth Embodiment Referring to FIG. 77, in the twenty-eighth embodiment, a data bus initialization unit 90 is added to the configuration of the twenty-seventh embodiment. The data bus initialization unit 90 is controlled by a reset signal PR from the access counter 50. The role of the data bus initialization unit (INIT) 90 is to set the internal data bus 24 by setting the two bus lines of each pair of complementary data bus lines to the same potential between the power supply potential and the ground potential. It is to initialize. This is accomplished by temporarily interconnecting the two bus lines, by precharging the bus lines to the desired intermediate potential, or by performing both interconnection and precharging together. The result is that each pair of complementary data bus lines is set to an intermediate level between a binary 1 level and a binary 0 level.

  The operation of the twenty-eighth embodiment is shown in FIG. The waveform shown here is generally the same as the waveform of FIG. 71 except that a PR waveform is added. The data bus initialization operation will be described below. Other operations are the same as those in the twenty-seventh embodiment.

  During the transfer of data from the memory array 2 to the buffer circuit 83 via the data bus 24 from time t2 to time t7, when the PW control signal is high, the PR control signal remains low. Yes, the data bus initialization unit 90 remains inactive.

  When the access counter 50 sets the PW signal to low level after time t7, the access counter 50 simultaneously sets the PR signal to high level for one clock cycle centered on time t8. During this clock cycle, the data bus 24 is initialized by the data bus initialization unit 90. The bit line in column Yj-4 (still connected to data bus 24 and memory cell Nij-4 in the same column) is also initialized. The data bus initialization unit 90 is deactivated at the timing between times t8 and t9, and the data bus 24 and these bit lines and memory cells are initialized.

  At time t9, new input data Daj-4 is received and transferred from the data input unit 14 to the data bus 24. The data bus line and the bit line can be quickly changed from the intermediate potential to a level representing binary 1 or binary 0, and the capacitor in the memory cell Nij-4 is quickly charged or discharged. Remember. In this way, data can be written in a short time, and the start of the next burst can be accelerated.

  If the data bus 24 is not initialized and the new input data Daj-4 is different from the old data Dbj-4, the data bus line and the bit line must change from one to the other between the power supply potential and the ground potential, The capacitor in the memory cell must be fully charged or discharged, so the time taken for the write operation is long and the time between bursts must be long.

  By shortening the time between bursts, the memory device of the twenty-eighth embodiment can output a longer burst.

Twenty-ninth Embodiment Referring to FIG. 79, the twenty-ninth embodiment is obtained by adding the address register 52 described in the fifth embodiment to the configuration of the twenty-seventh embodiment. The address register 52 receives and stores the Y address YAD output from the address input unit 81. As a result, the Y address YAD can be repeatedly loaded into the down counter 30. As described in the fifth embodiment, the Y address YAD is reloaded according to the external control signal ADX /.

  By storing the Y address YAD in the address register 52 instead of the address input unit 81, the Y address can be held near the down counter 30 in terms of the circuit configuration, and the address input unit 81 can be stored in the next Y address. You can start preparing for the reception.

  Since the down counter 30 does not need to be reloaded in the middle of a burst, a switch between the address register 52 and the down counter 30 is not necessarily required. The down counter 30 can be controlled by a PW control signal. That is, every time the PW control signal rises, the down counter 30 loads the Y address value held in the address register 52 and counts down from that value. The address register 52 is configured using, for example, a transparent latch, and the new address value received by the address input unit 81 can be immediately used for the down counter 30.

  The operation of the twenty-ninth embodiment will be understood from the description of the operations of the fifth embodiment and the twenty-seventh embodiment. Therefore, detailed description is omitted. However, a related description is given in the thirty-first embodiment. The advantage of the twenty-ninth embodiment is that it is not necessary to repeatedly input the same column address from the outside.

Thirty Embodiment Referring to FIG. 80, the thirtieth embodiment combines the features of the twenty-eighth embodiment and the twenty-ninth embodiment, and initializes the address register 52 and the data bus. Both units 90 are provided. Detailed description is omitted. The operation of the thirtieth embodiment is substantially the same as the operation of the next embodiment.

Thirty-first embodiment Referring to FIG. 81, the thirty-first embodiment is connected to the configuration of the thirtieth embodiment between the address register 52 and the down counter 30, and the control signal PO from the access counter 50 The address output switch 54 described in the fifth embodiment is added. Further, a control signal PM supplied from the access counter 50 to the address register 52 is added.

  FIG. 82 shows operations of the thirty-first embodiment. Hereinafter, the reloading operation of the down counter 30 will be described. Other operations are as described in the twenty-eighth embodiment.

  During the first burst from time t 1 to time t 6 in FIG. 82, the PO control signal remains low, preventing the down counter from being loaded from the address register 52. Instead, at time t2, the input Y address (Yj) received by the address input unit 81 is loaded directly into the down counter 30 via a signal line not shown in FIG. The Y address Yj is stored in the address register 52 in accordance with the control signal PM (not shown in FIG. 82) at any convenient time between time t2 and time t6.

  The first burst is performed in the same manner as described in the twenty-seventh and twenty-eighth embodiments. When the PR control signal is activated at time t4, the data bus 24 is initialized by the data bus initialization unit 90. This is because new data Daj-4 is written at time t5.

  The second burst begins with the input of a new X address (Xi + 1). At time t8, the CS /, CAS /, ADX / signals are at a low level, and when the PW control signal is driven to a high level, the access counter 50 sets the PO control signal to a high level. The PO control signal is high for one clock cycle. During this period, the address register output switch 54 is closed, and the same Y address (Yj) as before is loaded from the address register 52 to the down counter 30. During the second burst, the control signal PM (not shown) remains inactive and the address register 52 maintains the same address value. Therefore, the second burst accesses the next row (Xi + 1) data in the same column.

  In the thirty-first embodiment, the initialization of the data bus 24 brings about an improvement in the operating speed as in the twenty-eighth embodiment, and it is not necessary to repeatedly input the same column address. This embodiment has the same advantages as the first embodiment.

Compared with the twenty-ninth and thirty-first embodiments, the thirty-first embodiment does not require the address register 52 to immediately transmit the newly received address data to the down counter 30, so the design of the address register 52 Since the address register output switch 54 prevents the down counter 30 from receiving an undesirable address input from the address register 52 when no address input is required, the design of the down counter 30 is designed. Great freedom.

Thirty-second embodiment Referring to FIG. 83, the thirty-second embodiment is obtained by adding a block selection unit 92 for selecting different data blocks in the memory array 2 to the configuration of the thirty-first embodiment. It is.

  The address input unit 81 of this embodiment divides the received Y address bits into the upper group PC and the lower group PB. The lower group PB is supplied to the address register 52 and the down counter 30. The upper group PC is supplied to the block selection unit 92. The block selection unit 92 generates an upper Y address YUAD from the supplied address bits, and this upper Y address YUAD is directly supplied to the Y decoder 8 bypassing the down counter 30. The Y decoder 8 uses YUAD, for example, as the upper bits of the Y address.

  The memory control signal generator 68 according to the thirty-second embodiment supplies the control signal NBL to the access counter 50. This control signal NBL specifies the number of columns to be accessed in the same burst in each block. The access counter 50 supplies the next block control signal PNBL to the block selection unit 92, whereby the block selection unit 92 outputs the upper Y address YUAD for the next block.

  FIG. 84 shows an example of the internal configuration of the down counter 30, the address register 52, the address register output switch 54, and the block selection unit 92. The character “n” is the total number of Y address bits including upper address bits and lower address bits.

  The down counter 30 is composed of a series of 1-bit counters C0 to Cn-3 which are interconnected and driven by a counter clock signal CCLK. The counter clock signal CCLK is obtained by combining the internal clock signal CLK 'and the control signal PW with a NAND gate and an inverter as shown. Each 1-bit counter Ci is, for example, a circuit having an output that is inverted by a specific transition (rising or falling) of a signal output by the left-side 1-bit counter Ci-1. All output transitions are synchronized with the counter clock CCLK.

  The address register 52 includes latches E0 to En-3 and transistors Trdd0 to Trddn-3. The transistors Trdd0 to Trddn-3 are controlled by the PM control signal, and supply n-2 lower Y address bits PB (Y0 to Yn-3) from the address input unit 81 to the latches E0 to En-3. To do.

  The address register output switch 54 includes transistors Trd0 to Trdn-3 controlled by a PO control signal, and outputs the latches E0 to En-3 in the address register 52 to the corresponding 1-bit counters C0 to C0 in the down counter 30. Supply to Cn-3.

The block selection unit 92 includes a pair of 1-bit counters F0 and F1, a pair of 1-bit latches D0 and D1, a pair of transistors Trddn-2 and Trddn-1, and another pair of transistors Trdn-2, Trdn-1. The transistors Trddn-2 and Trddn-1 are controlled by the PM control signal, and two upper address bits Yn-2 and Yn-1 (PC) are supplied from the address input unit 81 to the 1-bit counters F0 and F1. . Transistors Trdn-2 and Trdn-1 are controlled by a PO control signal and supply the outputs of 1-bit counters F0 and F1 to latches D0 and D1, respectively. The 1-bit counters F0 and F1 are driven by a PNBL control signal and are interconnected to operate as an up counter or a down counter.

  The outputs SY0 to SYn-1 of the down counter 30 and the block selection unit 92 form a complete Y address signal supplied to the Y decoder 8. A value (YADD) combining the lower bits SY0 to SYn-3 is counted down in synchronization with the counter clock CCLK, and a value (YUAD) combining the upper bits SYn-2 and SYn-1 is counted in synchronization with the PNBL control signal. Up and down.

  Referring to FIG. 85, upper address bit YUAD output by block selection unit 92 divides memory array 2 into a plurality of blocks. For simplicity, only one upper address bit and two blocks (block a and block b) are shown in FIG. The value of the upper address bit (YUAD) of the block a is 0, and the value of the upper address bit of the block b is 1.

  The memory of FIG. 85 can be used to store one field of pixel data in each block, and therefore the memory array 2 can hold two fields or one frame of pixel data. For example, even fields are stored in block a and odd fields are stored in block b.

  A single burst contains data from both fields. FIG. 85 shows the case where data is currently being received for the even field and is being stored in block a. The burst starts with reading the data (Yb3, Yb2, Yb1) of the previous odd field from the block b, and then the data (Yc3, Yc2) of the previous even field from the block a. , Yc1) is read out, and then new input data (Ya1 not shown) is overwritten (indicated by hatched dots) on the oldest read data (Yc1).

  In this burst, after the transfer of the data Yb1, the upper address bit changes from 1 to 0, and the lower address bit of the start address (“10111”) is loaded from the address register 52 to the down counter 30 again. It is obtained by controlling the block selection unit 92 and the address register output switch 54 with PNBL and PO control signals. Therefore, the burst jumps without interruption from the column address “110101” (Yb1) in the block b to the column address “010111” (Yc3) in the block a.

  When all the new even field data is stored in block a during reception of the next field (odd field) data, burst access is performed as shown in FIG. This time, each burst begins in block a and ends in block b, and new pixel data is stored in block b.

  When sequential scanning is used, different frames of data are stored in each block rather than different fields. If different blocks store pixel data for different frames, the data represents the pixel at the same position in different frames. If different blocks store different fields of pixel data, the data represents pixels in adjacent positions in different fields. For example, as shown in FIG. 64, data Da1 and Db1 represent pixels at positions adjacent to each other in fields a and b.

  According to the thirty-second embodiment, a single memory device can output pixel data from a plurality of fields or frames in a single burst. 85 and 86, the number of fields or frames is two. However, according to the number of upper address bits, the memory array 2 of the thirty-second embodiment can be divided into an arbitrary number of blocks each storing different fields or frames. Unlike the first to twenty-sixth embodiments, the thirty-second embodiment does not require data to be transferred from one block to another. Therefore, the operation is simpler than the previous embodiment. However, the amount of data to be stored is large.

  Further details of the thirty-second embodiment will be described in connection with the thirty-eighth embodiment.

Thirty-third embodiment Referring to FIG. 87, the thirty-third embodiment is similar to the twenty-seventh embodiment, but has two memory banks, separate memory arrays 2A and 2B, and separate X Decoders 6A and 6B and separate Y decoders 8A and 8B are provided. Both banks share the same internal data bus 24 and down counter 30.

  One of the two banks stores even X address pixel data and the other bank stores odd X field pixel data of the same field, thus precharging one bank during access to the other bank can do.

  Details of the operation of the thirty-third embodiment will be understood from the description of the thirty-eighth embodiment.

Thirty-fourth embodiment Referring to FIG. 88, the thirty-fourth embodiment combines the features of the twenty-eighth embodiment and the features of the thirty-third embodiment. That is, in the thirty-fourth embodiment, a data bus initialization unit 90 is added to the two-bank configuration of the thirty-third embodiment. The thirty-fourth embodiment combines the advantages of the twenty-eighth embodiment and the thirty-third embodiment.

  Details of the operation of the thirty-fourth embodiment will be understood from the description of the thirty-eighth embodiment.

Thirty-fifth Embodiment Referring to FIG. 89, the thirty-fifth embodiment is a combination of the features of the twenty-ninth embodiment and the features of the thirty-third embodiment. That is, the thirty-fifth embodiment is obtained by adding an address register 52 to the configuration of the thirty-third embodiment. The thirty-fifth embodiment has the advantages of the twenty-ninth embodiment and the thirty-third embodiment.

  Details of the operation of the thirty-fifth embodiment will be understood from the description of the thirty-eighth embodiment.

Thirty-sixth embodiment Referring to FIG. 90, the thirty-sixth embodiment is a combination of the features of the thirty-third embodiment and the features of the thirty-third embodiment. That is, in the thirty-sixth embodiment, an address register 52 and a data bus initialization unit 90 are added to the configuration of the thirty-third embodiment. The thirty-sixth embodiment combines the advantages of the thirty-third embodiment and the thirty-third embodiment.

  Details of the operation of the thirty-sixth embodiment will be understood from the description of the thirty-eighth embodiment.

Thirty-Seventh Embodiment Referring to FIG. 91, the thirty-seventh embodiment combines the features of the thirty-first embodiment and the thirty-third embodiment. That is, the thirty-seventh embodiment is obtained by adding an address register output switch 54 to the configuration of the thirty-sixth embodiment. The thirty-seventh embodiment combines the advantages of the thirty-first embodiment and the thirty-third embodiment.

  Details of the operation of the thirty-seventh embodiment will be understood from the description of the thirty-eighth embodiment.

Thirty-eighth embodiment Referring to FIG. 92, the thirty-eighth embodiment is a combination of the features of the thirty-second and thirty-third embodiments, and is the same as the thirty-second embodiment. It has a block selection unit 92 and two memory banks as in the thirty-third embodiment. Each memory bank is divided into a plurality of blocks according to the upper address bit YUAD output by the block selection unit 92.

  FIG. 93 shows the internal configuration of the thirty-eighth embodiment in more detail.

  The address input unit 81 outputs an X address signal XAD and three other control and address signals PA, PB, PC. PA is supplied to the access counter 50 and controls the burst length. This point is as described in the twenty-seventh embodiment. PB is composed of lower bits of the Y address and is supplied to the address register 52. The PC consists of the upper bits of the Y address and is supplied to the block selection unit 92.

  Access counter 50 receives an NBL control signal from memory control signal generator 68. As described in the thirty-second embodiment, the NBL controls the number of bits read from each block by one burst. The access counter 50 outputs a PO control signal as shown in the thirty-first embodiment, and outputs a PNBL signal for controlling the block selection unit 92 as described in the thirty-second embodiment. A flag signal is output that is at a low level and becomes a high level when indicating the end of a burst.

  The address register 52, the address register output switch 54, and the down counter 30 have, for example, the configuration shown in FIG. However, the address register output switch 54 receives a PO control signal from the access counter 50 and a NO control signal from the memory control signal generator 68. The address register output switch 54 connects the address register 52 to the down counter 30 when NO or PO is activated.

  The memory control signal generator 68 outputs various control signals in addition to those described in the previous embodiment. Two of them, the output enable signal POE and the read timing signal PTR are clearly shown. Other control signals are collectively indicated by thick arrows. The memory control signal generator 68 receives the flag signal output from the access counter 50.

  The block selection unit 92 has a structure shown in FIG. 84, for example, and has 1-bit counters F0 and F1 for generating the upper Y address signal YUAD.

  The internal data bus 24 has a pair of complementary signal lines Da and Da / for the bank A and another pair of complementary signal lines Db and Db / for the bank B. For simplicity, only one pair of data bus lines is shown for each bank. The data bus 24 is also read bus lines RD, RD / and RRDa, RRDa / (by which the data bus lines Da, Da /, Db, Db / are coupled to the buffer circuit 83 via transistor switches (described later)). And the write data bus lines WDa and WDa / (by which the data bus lines Da and Da / and Db and Db / are coupled to the data input unit 14 via other transistor switches (described later)). Of complementary pairs.

  The data bus initialization unit 90 includes transistors Tra1 and Trb1 that equalize complementary pairs of data bus lines. These transistors are driven by reset control signals PRa and PRb (a part of control signals generated by the memory control signal generator 68).

  Data bus lines Da and Da / drawn from bank A are coupled to write data bus lines WDa and WDa / through transistor SWa. Transistor SWa is driven by internal write enable control signal PWEa for bank A. Similarly, the data bus lines Db and Db / drawn from the bank B are coupled to the write data bus lines WDa and WDa / via the transistor SWb. Transistor SWb is driven by internal write enable control signal PWEb for bank b. The control signals PWEa and PWEB are output from the memory control signal generator 68. Write data bus lines WDa and WDa / are coupled to data input unit 14.

  Data bus switch 82 that couples data bus 24 to buffer circuit 83 includes transistors SRa1, SRb1, and SR2. Transistor SRa1 couples data bus lines Da, Da / of bank A to read data bus lines RD, RD / and is driven by an internal read enable control signal PREa for bank A. Transistor SRb1 couples data bus lines Db, Db / of bank B to read data bus lines RD, RD / and is driven by an internal read enable control signal PREb for bank B. The transistor SR2 couples the read data bus lines RD and RD / to the read data bus lines RRDa and RRDa /, and is driven by a read ready (ready) control signal PRR output from the memory control signal generator 68. . Read data bus lines RRDa and RRDa / are coupled to buffer circuit 83. An amplifier 96 is coupled to the read data bus lines RD and RD / and the data bus lines Db and Db / (from the transistor SRb1 to the memory array 2B) to amplify the read data.

  The read enable control signals PREa and PREb include, for example, AND gates and OR gates, and are generated by a logic circuit 98 that receives a PTR control signal from the memory control signal generator 68 and a flag signal from the access counter 50. Logic circuit 98 is also output by memory control signal generator 68 and receives control signals Pa and Pb for selecting banks A and B, respectively.

  The data output unit 16 is controlled by an internal output enable signal POE from the memory control signal generator 68.

  The Y decoders 8A and 8B have AND gates as shown in FIG. In FIG. 93, these AND gates in the Y decoder 8A are indicated by YDa1 to YDan, and those in the Y decoder 8B are indicated by YDb1 to YDbn.

  FIG. 94 shows the internal structure of the memory arrays 2A and 2B. This structure is the same as that of the main memory array 2 in FIG. The following description relates to the difference in symbols.

  93, YDa1 (YDb1) to YDan (YDbn) represent AND gates in the Y decoders 8A and 8B, and these are equivalent to the AND gate 38 in FIG.

  The signals indicated by Y1 to Ym in FIG. 10 are indicated by Ya1 (Yb1) to Yan (Ybn) in FIG. The letters “a” and “b” represent banks A and B, respectively.

  The transfer transistors 46 that connect the bit lines to the data bus lines Da (Db) and Da / (Db /) are denoted by reference numerals S1 to Sn. The bit lines are indicated by BL1, BL1 / to BLn, BLn /.

  PSAA and PSAB are signals for activating the sense amplifiers (SA) of the banks A and B. T1 to Tn represent columns 1 to n. The letter “m” represents the number of rows in each memory bank. The letter “n” represents the number of columns in each memory bank.

  Next, the operation of the thirty-eighth embodiment will be described. The following description is for a burst where 6 pixels of data are read from one bank, 6 pixels are taken from 3 fields each from 2 fields, and the data is stored in separate banks. Stored in block.

  The following description also applies to the thirty-second to thirty-seventh embodiments as long as these embodiments have the same features as the present embodiment.

  Referring to FIG. 95, at time t1, CS /, RAS / are at the low level, and the X address (Xi) is received by the address input terminal ADD. The memory control signal generator 68 selects the bank A or B, for example, according to the LSB of the X address, and sends the X address to the corresponding X decoder 6A, 6B, which drives the corresponding word line WLi. . In the following description, it is assumed that bank A is selected. In this case, the control signal Pa is at a high level and the control signal Pb is at a low level.

  After time t1, the memory control signal generator 68 drives the sense amplifier activation signal (PSAA) in the selected bank A (to high level), and the bit lines (BL1, BL1 / ˜BLn in the bank A). , BLn /), data begins to appear. Bank B remains inactive.

  At time t2, CS /, CAS / are at a low level, and the Y address Yj is received at the address input terminal ADD /. The address input unit 81 sends the lower address bits (PB) to the address register 52 and sends the upper address bits (PC) to the block selection unit 92. The address input unit 81 also sends a PA control signal to the access counter 50 to specify a burst length of 6 cycles. The NBL control signal sent from the memory control signal generator 68 to the access counter 50 indicates that the burst should spend 3 cycles accessing each block in the selected memory bank.

  The present invention is not limited to a particular method using PA and NBL control signals. As another method, PA may specify 3 cycles, and NBL may specify that each 3 cycles is repeated twice.

  At a time between times t2 and t3, the memory control signal generator 68 activates the NO control signal, and the address register output switch 54 connects the address register 52 to the down counter 30 in response to this, and the lower Y address bit Is loaded into the down counter 30. These bits are combined with the upper address bit YUAD output output from the block selection unit 92, and the Y decoder 8A receives the entire input Y address Yj and transfers the bit line in the designated column Tj in the bank A to the data The signal Yaj connected to the bus 24 is activated.

  Note that the characters “a” and “b” in the decoder output waveform (Yaj, Yaj−1,..., Ybj−2) in FIG. All accesses of this burst are for bank A. “A” and “b” in the decoder output waveform indicate blocks a and b in the bank A as shown in FIG.

  At time t3, data Daj (corresponding to data Da3 in FIG. 86) is transferred from bit lines BLj and BLj / and starts appearing on data bus lines Da and Da / in bank A. During the next clock cycle, the internal read enable control signal PREa is activated for a certain period, during which data Daj is transferred from the data bus lines Da, Da / to the read data bus lines RDa, RDa / via the transistor SRa1. (See waveform in FIG. 95). The PRR control signal (not shown) is also activated, and the data Daj is transferred to the buffer circuit 83 via the transistor SR2 and the read data bus lines RRDa and RRDa /, latched in the buffer circuit 83, and data output It is transmitted to the unit 16.

  The PREa signal is generated from the PRR read timing signal by the logic circuit 98 shown in FIG. The flag signal output by the access counter 50 is at a low level, and Pa is at a high level. Thus, PREa follows the high and low level transitions of the PTR (generally as shown in the waveform of FIG. 95).

  A little later than time t3, the memory control signal generator 68 activates the output enable control signal POE, and in the next cycle, the data output unit 16 outputs data Daj to the output terminal DOUT as shown.

  In this way, the operation proceeds (while the count value of the down counter 30 decreases by 1), and the data Daj-1 and Daj-2 are transferred from the memory array 2A to the buffer circuit 83. These transfers are performed in clock cycles before and after time t4.

  A little later than time t5, the access counter 50 determines from the information previously supplied by the PA and NBL signals that sufficient data has been read from the block a, activates the PO control signal, and starts the start Y address (lower bit). ) Is reloaded from the address register 52 to the down counter 30. At the same time, although not shown in FIG. 95, the access counter 50 sends a PNBL control signal to the block selection unit 92 and changes the upper address bit YUAD to the content indicating the next block (b). Thus, at time t5, the Y decoder 8A generates the signal Ybj for selecting the column in the block b in the memory bank A, and the data Dbj (corresponding to Yb3 in FIG. 86) is the data bus lines Da, Da /, The data is transferred to the buffer circuit 83 via RDa, RDa /, RRDa, and RRDa /.

  After times t6 and t7, subsequent data Dbj-1 and Dbj-2 (corresponding to Yb2 and Yb1 in FIG. 86) are transferred from the memory array 2A to the data bus lines Da, Da /, RDa, RDa /, RRDa, and RRDa. The data is transferred to the buffer circuit 83 via /. Data Dbj, Dbj-1, and Dbj-2 are output by data output unit 16 at times t6, t7, and t8, respectively.

  At time t8, when the data Dbj-2 has been transferred to the buffer circuit 83 and is being output by the data output unit 16, the memory control signal generator 68 activates the data bus reset control signal PRa, and the data The bus lines Da and Da / are initialized. The Y decoder 8A continues to output the Ybj-2 column selection signal, so that the bit line selected by this signal is also initialized. New input data (indicated by hatched dots) is received at the data input terminal DIN at time t8 and supplied to the write data bus lines WDa and WDa /.

  At this time, the access counter 50 activates the flag signal to indicate the end of the burst. Upon receipt of the flag signal, the memory control signal generator 68 stops outputting the read timing signal PTR, and therefore no PTR pulse is generated after time t8. Therefore, the PREa pulse is not generated after time t8.

  In FIG. 93, the logic circuit 98 that generates the read enable signals PREa and PREb is shown as receiving a flag signal. However, PREa and PREb can be output without inputting the flag signal to the logic circuit 98. This input can be saved because it can be generated correctly.

  A little later than time t8, the memory control signal generator 68 activates the internal write enable signal PWEa and connects the write data bus lines WDa and WDa / to the data bus lines Da and Da / for the memory array 2A. At time t9, new data is transferred to the data bus lines Da and Da /. Since the Y decoder 8A continues to output the signal Ybj-2, the input data is transferred to the selected bit line and written into the memory cell in the block that has previously stored the data Dbj-2. These bit lines are shown as B1j and BLj / in FIG. 95, but are not the same as the same bit lines BLj and BLj / from which the first data Daj in the burst was read. This is because the data Daj is read from the block a.

  During this burst, the word lines in bank B can also be precharged in preparation for the next burst (bank B is accessed). Data bus lines and bit lines in bank A can also be initialized.

  In FIG. 95, the flag signal is shown to remain high from time t8 to time t12. However, this flag signal may be set to a low level before that to prepare for the next burst. Other minor variations on timing are possible, and FIG. 95 does not show the exact timing relationship, but only shows a rough flow of each event.

  Similar to the thirty-second embodiment, according to the thirty-eighth embodiment, pixel data from a plurality of fields or frames can be output in a single burst by a single memory device. Furthermore, according to the thirty-eighth embodiment, the interval between bursts can be shortened by hiding the precharging of each bank from the access of other banks (by performing it simultaneously). By using two banks, in the thirty-eighth embodiment, the preferred use of the thirty-eighth embodiment is to store data for odd-numbered pixels in each scan line in one bank, The data for even-numbered pixels in the scan line is stored in another bank and the two banks are accessed alternately.

Modified Example In the above embodiment, bank interleaving is performed in which every other row is accessed in an alternate bank. However, column interleaving can also be performed between banks.

  The number of banks is not limited to two and may be larger.

  An up counter may be used in place of the down counter in the Y address generator. The advantage of using a down counter is that scan lines in a moving image are generally addressed and scanned in ascending order from the top to the bottom of the screen. In this case, the down counter is more convenient.

  When the first to twenty-sixth embodiments are used in the non-dependent connection configuration and the input and output are performed at different times, the data input unit 14 and the data output unit 16 can share the same external data terminal.

  The access count register and the address recalculator of the nineteenth to twenty-sixth embodiments can be used in other embodiments.

  The PM and PO control signals of the thirty-first embodiment may be generated not by the access counter 50 but by the memory control signal generator 68.

  The PA control signal of the twenty-seventh embodiment may be generated not by the address input unit 81 but by the memory control signal generator 68.

  Although the present invention has been described as solving the problem in the digital processing of moving images, the memory device of the present invention can be used for applications other than digital processing of moving images. Those skilled in the art will appreciate that further variations are possible within the scope of the appended claims.

It is a figure which shows the group of a pixel. It is a figure which shows the group of the pixels in the several continuous field of a moving image. It is a figure which shows the group of many pixels in the continuous field of a moving image. FIG. 4 illustrates a conventional system for read access to the pixel of FIG. It is a figure which shows an example of a read and write access. It is a figure which shows the other example which shows read and write access. FIG. 6 is a diagram illustrating a part of FIG. 4 corresponding to the access operation of FIG. 5. FIG. 7 is a diagram showing a part of FIG. 4 corresponding to the access operation of FIG. 6. It is a block diagram which shows the 1st Embodiment of this invention. It is a circuit diagram which shows 1st Embodiment in detail. FIG. 4 is a timing diagram showing a read access to the main memory array according to the first embodiment. FIG. 6 is a timing diagram illustrating a read access to the sub memory array according to the first embodiment. It is a circuit diagram which shows the 2nd Embodiment of this invention. It is a circuit diagram which shows the 3rd Embodiment of this invention. It is a circuit diagram which shows the 4th Embodiment of this invention. It is a figure which shows the usage method of the main and sub memory array of 1st thru | or 4th Embodiment. It is a circuit diagram which shows the 5th Embodiment of this invention. FIG. 20 is a timing diagram illustrating a burst access operation according to the fifth embodiment. It is a figure which shows a part of FIG. 4 corresponding to the access operation | movement of FIG. It is a timing diagram which shows reload of the down counter of 5th Embodiment. It is a circuit diagram which shows the 6th Embodiment of this invention. It is a circuit diagram which shows the 7th Embodiment of this invention. It is a circuit diagram which shows the 8th Embodiment of this invention. It is a circuit diagram which shows the 9th Embodiment of this invention. It is a figure which shows a part of system of FIG. 4 which can be replaced by 9th Embodiment. It is a conceptual diagram which shows storage and access of the pixel data in 9th Embodiment. It is a conceptual diagram which shows storage and access of the pixel data in 9th Embodiment. It is a conceptual diagram which shows storage and access of the pixel data in 9th Embodiment. It is a conceptual diagram which shows storage and access of the pixel data in 9th Embodiment. It is a figure which shows the pixel data accessed by the single burst in 9th Embodiment. FIG. 31 is a timing diagram showing a burst for accessing the pixel data shown in FIG. 30 in a manner suitable for dependent connection. FIG. 31 is a timing diagram showing another burst for accessing the pixel data shown in FIG. 30. It is a figure which shows more pixel data accessed by the single burst in 9th Embodiment. FIG. 34 is a timing diagram showing a burst for accessing the pixel data shown in FIG. 33 in a manner suitable for slave connection. FIG. 34 is a timing diagram showing another burst for accessing the pixel data shown in FIG. 33. It is a figure which shows the still more pixels accessed by the single burst in 9th Embodiment. FIG. 37 is a timing chart showing a burst for accessing the pixel data shown in FIG. 36. It is a block diagram which shows the 10th Embodiment of this invention. It is a block diagram which shows the 11th Embodiment of this invention. It is a block diagram which shows the 12th Embodiment of this invention. It is a block diagram which shows the 13th Embodiment of this invention. It is a figure which shows operation | movement of 13th Embodiment in non-subordinate connection mode. It is a figure which shows operation | movement of 13th Embodiment in non-subordinate connection mode. It is a figure which shows operation | movement of 13th Embodiment in non-subordinate connection mode. It is a figure which shows operation | movement of 13th Embodiment in non-subordinate connection mode. It is a figure which shows operation | movement of 13th Embodiment in non-subordinate connection mode. It is a figure which shows operation | movement of 13th Embodiment in non-subordinate connection mode. It is a figure which shows operation | movement of 13th Embodiment in the beginning of a burst. It is a block diagram which shows the 14th Embodiment of this invention. It is a block diagram which shows the 15th Embodiment of this invention. It is a block diagram which shows the 16th Embodiment of this invention. It is a block diagram which shows the 17th Embodiment of this invention. It is a block diagram which shows the 18th Embodiment of this invention. It is a block diagram which shows the 19th Embodiment of this invention. It is a block diagram which shows the 20th Embodiment of this invention. It is a block diagram which shows the 21st Embodiment of this invention. It is a block diagram which shows the 22nd Embodiment of this invention. It is a block diagram which shows the 23rd Embodiment of this invention. It is a block diagram which shows the 24th Embodiment of this invention. It is a block diagram which shows the 25th Embodiment of this invention. It is a block diagram which shows the 26th Embodiment of this invention. It is a figure which shows the pixel data accessed in the filtering operation | movement performed with respect to a continuous field using a different number of pixels from an even field and an odd field. It is a figure which shows the pixel data accessed in the filtering operation | movement performed with respect to a continuous field using a different number of pixels from an even field and an odd field. It is a figure which shows the pixel data accessed in the filtering operation | movement performed with respect to a continuous field using a different number of pixels from an even field and an odd field. FIG. 62 is a timing chart showing burst access to the pixel data shown in FIG. 62 in a twenty-sixth embodiment. FIG. 67 is a timing chart showing burst access to the pixel data shown in FIG. 64 in the twenty-sixth embodiment. It is a figure which shows the modification of previous embodiment. It is a block diagram which shows the 27th Embodiment of this invention. FIG. 69 is a more detailed block diagram of the address input unit of FIG. 68. FIG. 69 is a more detailed block diagram of the access counter of FIG. 68. FIG. 38 is a timing diagram illustrating the operation of the twenty-seventh embodiment, showing dependently connected inputs and outputs. It is a figure which shows arrangement | positioning of the field data in the memory array in 27th Embodiment. It is a figure which shows the continuous burst access by 27th Embodiment. It is a figure which shows the continuous burst access by 27th Embodiment. It is a figure which shows the circuit which receives the data output by 27th Embodiment. FIG. 38 is a diagram illustrating subordinate connections of the memory device according to the twenty-seventh embodiment. It is a block diagram which shows the 28th Embodiment of this invention. FIG. 44 is a timing chart showing the operation of the twenty-eighth embodiment. . It is a block diagram which shows the 29th Embodiment of this invention. It is a block diagram which shows the 30th Embodiment of this invention. It is a block diagram which shows the 31st Embodiment of this invention. FIG. 38 is a timing diagram showing an operation according to the thirty-first embodiment. It is a block diagram which shows the 32nd Embodiment of this invention. It is a figure which shows an example of the internal structure of the address register in 32nd Embodiment, an address register output switch, a down counter, and a block selection unit. It is a figure which shows an example of the data accessed by a single burst in 32nd Embodiment. FIG. 38 is a diagram illustrating another example of data accessed in a single burst in the thirty-second embodiment. It is a block diagram which shows the 33rd Embodiment of this invention. It is a block diagram which shows the 34th Embodiment of this invention. It is a block diagram which shows the 35th Embodiment of this invention. It is a block diagram which shows the 36th Embodiment of this invention. It is a block diagram which shows the 37th Embodiment of this invention. It is a block diagram which shows the 38th Embodiment of this invention. It is a more detailed block diagram of the thirty-eighth embodiment. FIG. 38 is a conceptual diagram showing a memory array in the thirty-eighth embodiment. FIG. 44 is a timing chart showing the operation of the thirty-eighth embodiment.

Explanation of symbols

  A, B memory bank, 2, 2A, 2B main memory array, 4, 4A, 4B sub memory array, 6, 6A, 6B row decoder, 8, 8A, 8B main column decoder, 10, 10A, 10B sub column decoder, 12, 12A, 12B column address generator, 14 data input unit, 16 data output unit, 18 upper address bus, 20 lower address bus, 22 address bus switch, 24, 24A, 24B internal data bus, 26 data bus switch, 28 Write amplifier, 30 address counter, 40 address holding latch, 47 main address bus, 48 sub address bus, 50 access counter, 52 address register, 54 address register output switch, 56 bank bus switch, 62, 62A Send register, 64 input data register, 68 memory control generator, 70, 72, 74 access count register, 76A, 76B address recalculator, 82 data bus switch, 83 buffer circuit, 90 data bus initialization unit, 92 blocks Selection unit.

Claims (8)

A memory device having two memory banks for receiving a row and column address signal, input data, and an external control signal in synchronization with a clock signal,
A control signal generator ( 68 );
A data input unit (14) for receiving the input data;
A data output unit (16) for outputting output data;
A bank bus switch (56) and a transfer register (62) coupled to the bank bus switch and shared by all of the two memory banks;
Each memory bank (A, B) has a plurality of word lines,
A row decoder (6A, 6B) for activating a word line selected by the received row address signal among the word lines;
A main memory array (2A, 2B);
A column address generator (12A, 12B);
A main column decoder (8A, 8B);
A sub memory array (4A, 4B);
A sub-column decoder (10A, 10B);
Internal data buses (24A, 24B) and
The bank bus switch (56) couples the internal data bus (24A, 24B) to the data input unit (14) and the data output unit (16),
The main memory array (2A, 2B) has a plurality of memory cells arranged to form mutually intersecting rows and columns, and the word line is coupled to each row of the memory cells;
The column address generator (12A, 12B) generates a series of column addresses from a single received column address signal, each of the column addresses having an upper portion and a lower portion;
The main column decoder (8A, 8B) is coupled to the main memory array and the column address generator, decodes the column address, and transfers a column of the corresponding memory cell in the main memory array to the internal data bus ( 24A, 24B)
The sub memory array (4A, 4B) has a plurality of memory cells arranged so as to form mutually intersecting rows and columns, and the word line is also provided for each column of memory cells in the sub memory array. The number of columns in the secondary memory array is less than the number of columns in the main memory array;
The sub column decoder (10A, 10B) is coupled to the sub memory array and the column address generator, decodes a lower part of the column address, and selects a corresponding column of memory cells in the sub memory array as the internal memory. Coupled to the data bus,
The control signal generator (68)
Coupled to the column address generator, receives the external control signal, and based on this, enables the main column decoder and the sub column decoder, and generates an internal control signal for controlling the data input unit and the data output unit. And this
Input new input data into the data input unit (14),
Data stored in a plurality of columns of the main memory array (2A) of one of the two memory banks is read, and the internal data bus (24A ) and the bank of the one memory bank are read The data transferred through the bus switch (56) , old data read from one column of the plurality of columns transferred in this way is temporarily stored in the transfer register (62), and the data New input data input to the input unit (14) and new data read from columns other than the one column among the transferred columns of data are transferred to the data output unit (16). through by output,
Data stored in a plurality of columns of the main memory array ( 2B) of the other memory bank of the two memory banks is transferred to the internal data bus ( 24B) of the other memory bank, the bank bus switch (56) and between the Ru is outputted through the output unit (16), a new input data input to said data input unit (14), the internal data bus of one of the memory banks described above (24A) The old data stored in the transfer register (62) is transferred from the transfer register (62) to the one column of the main memory array (2A) of the one memory bank. Transferred to the secondary memory array (4A),
The old data stored in the transfer register (62) is output via the data output unit (16), and the old data stored in the sub memory array (4A ) of the one memory bank is output. the internal data bus (24 a), the bank bus switch (56) and a memory device which Ru is output through the data output unit (16). An access coupled to the control signal generator (68) and storing a value provided from the control signal generator (68) indicating the number of column addresses to be generated within the series of column addresses in each bank. A count register (70);
An access counter (50) coupled to the access count register (70) for controlling a column address generator in each bank in accordance with a value stored in the access count register (70). 2. The memory device according to 1. A first access count register (70) coupled to the control signal generator (68) for storing a first value supplied from the control signal generator (68);
A second access count register (72) coupled to the control signal generator (68) and storing a second value supplied from the control signal generator (68);
A third access count register (74) coupled to the control signal generator (68) and storing a third value supplied from the control signal generator (68);
And an access counter,
The above access counter
Coupled to the first access count register, the second access count register, and the third access count register;
Controlling a column address generator in the one memory bank according to the first value, causing a corresponding amount of data to be output from the main memory array in the one memory bank;
Controlling a column address generator in the other memory bank in response to the second value, causing a corresponding amount of data to be output from the main memory array in the other memory bank;
The memory device according to claim 1, wherein a column address generator in the one memory bank is controlled in accordance with the third value, and a corresponding amount of data is output from a sub memory array in the one memory bank. .   Coupled to the column address generator in each bank and in response to a control signal supplied from the control signal generator (68), an initial of a series of column addresses generated by the respective column address generator 4. The memory device of claim 3, further comprising at least two address recalculators that modify column addresses, thereby modifying the amount of data output from the main memory array and sub memory array in each bank. . A method for controlling access to a memory array of a memory device, comprising:
Two banks are provided in the memory device,
Each bank has a main memory array (2A, 2B) and a sub memory array (4A, 4B),
Different banks have different rows of memory cells,
The memory array of each bank has rows and columns of memory cells intersecting each other;
The secondary memory array of each bank has the same number of rows as the main memory array and fewer columns than the main memory array;
(A) receiving a row address signal and activating corresponding rows in the main memory array ( 2A ) and the sub memory array ( 4A ) in the first bank of the two banks;
(A ′) receiving a row address signal and activating corresponding rows in the main memory array ( 2B ) and the sub memory array ( 4B ) in the second bank of the two banks;
(B) generating, within the memory device, a first series of column addresses designating consecutive columns in the main memory array ( 2A ) in the first bank;
(C) inputting new input data into the first bank;
(D) Data is read from the memory cell arranged at the intersection of the row in the first bank activated in step (a) and the column specified in step (b). Of the data , old data read from the one column in the main memory array is transferred to the transfer register, and the transferred old data is temporarily stored in the transfer register and input to the data input unit. Outputting new input data and new data read from a column other than one column in the main memory array ;
(E) generating a second series of column addresses specifying memory cells in successive columns in the main memory array (2B) in the second bank;
(F) Read and output data from the memory cell arranged at the intersection of the row in the second bank activated in the step (a ′) and the column designated in the step (e) In the meantime, new input data input to the data input unit is stored in the one column in the main memory array in the first bank, and stored in the transfer register (62) in the step (d). Transferring stored old data to the sub-memory array (4A) in the first bank;
(G) generating, in the memory device, a third series of column addresses designating consecutive columns in the sub-memory array ( 4A ) in the first bank;
(H) The old data stored in the transfer register (62) is output via the data output unit, and the row in the first bank activated in the step (a) and the step ( controlling access to the memory array ( 2A, 2B ) of the memory device comprising: reading old data from the memory cells located at the intersections of the columns specified in g) and outputting the read old data Method. The data stored in the memory array represents moving image pixels scanned along a plurality of scanning lines, and the pixel data of each scanning line corresponds to the scanning line in the main memory array in each bank. 6. The method of claim 5 , wherein the method is stored in a column. The data read in step (d), the data read in step (h), and data is read in step (f), according to claim represents the pixels in three different fields of said moving picture 6 The method described in 1. The data read in step (d), the data read in step (h), and data is read in step (f), according to claim represents the pixels in three different frames of said moving picture 6 The method described in 1.

Images