JP3304893B2 - Memory selection circuit and semiconductor memory device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a memory selection circuit, and more particularly to a memory selection circuit for performing continuous access by multi-bit prefetch.

[0002]

2. Description of the Related Art In recent years, the operating speed of a CPU has been remarkably improved. However, on the other hand, the operating speed of a main storage device such as a DRAM is slower than the operating speed of a CPU, so that a cache memory, which is a high-speed memory, is generally inserted between the CPU and the main storage device. When the cache memory is used, the rate at which the CPU accesses the cache memory (hit rate) can be up to about 90%, so the rate at which the CPU must access the main storage device (miss hit rate) is about 10%. Thus, it becomes possible to read the instruction at a very high speed.

On the other hand, such a high hit ratio is realized based on an empirical rule that an instruction stored near an address where an used instruction is stored has a high probability of being used immediately thereafter. This is because, after first reading the currently necessary instructions, the instructions in which only the lower several bits of the address are stored at different addresses are also read, and these are stored together in the cache memory. Therefore, when a cache memory is used, the main storage device accesses the data as described above, that is, first outputs necessary data, and then, stores data of an address having a different address only in the lower few bits of the address where the data is stored. Must be able to be output continuously.

Conventionally, such continuous access is performed by nibble mode access in a DRAM. The access of the 4MDRAM in the nibble mode will be described with reference to FIG.

FIG. 15 is a schematic diagram of a 4MDRAM performing 4-bit nibble mode access. In the figure, reference numeral 1500 denotes a 4M-bit memory cell array, and as shown in FIG.
And has been split into. Each block has 1024 bit lines and 1024 word lines, respectively, and the word line is common to each block. Reference numeral 1501 denotes a 10-bit column address (A0 to A9),
One of 024 column selection lines (hereinafter referred to as CSL)
Select a book, and then select the 10
One of the 24 bit lines is selected. 1
Reference numeral 502 denotes a 10-bit row address (A0 to A9) for selecting one of 1024 word lines commonly connected to each block. Reference numerals 1503 and 1504 denote a row address A10 and a column address A10 for determining an address to be accessed in the nibble mode, which are input to the nibble decoder 1505, respectively. Nibble decoder 1505 receives and decodes row address A10 and column address A10, selects one of the four output lines to be active, and receives the CAS clock to be active. Change the line selection one after another.

Next, the read operation of the 4MDRAM shown in FIG. 15 will be described. First, each block is given a column address 1501 and a row address 1502,
One CSL and one word line are selected. As a result, in each block, one bit line and one
Since one word line is selected, one memory cell is selected from each block. That is, by the column address 1501 and the row address 1502,
Four memory cells, each in the same address space, are simultaneously selected, and 4-bit data is simultaneously read. Each of these 4-bit data is latched by a data latch circuit. Next, the nibble decoder 1505
, One of the four output lines of the nibble decoder 1505 is selected and set to the active level. In response, one of the read 4-bit data is output. Thereafter, every time the CAS clock input to the nibble decoder changes, the output of the nibble decoder is changed one after another, and the read data is read in accordance with this.
Bit data is output one after another. That is, in the 4MDRAM shown in FIG. 15, one CSL is selected by an address excluding the lower two bits of an address given by the CPU (hereinafter, referred to as a CPU address), and the selected 1
After reading a total of four bits, one bit from each block, by the CSL of this book, the CPU address data is first output based on the column address A10 and the row address A10, which are the lower two bits of the CPU address, and then the CAS clock is output. This means that the data of the different addresses only in the lower two bits are successively output according to the change.

As described above, the DRAM that performs nibble mode access responds to the request of the cache memory, and it can be seen that the cache memory can be used if it is used as the main storage device.

[0008]

As described above, data can be continuously read out by nibble mode access. In recent years, however, data that is simultaneously output at one output timing (hereinafter, referred to as data width) is not available. Since the number of bits to be increased from the conventional one bit to four bits, eight bits, and sixteen bits, the nibble mode in which data to be output is read at one time requires a very large number of buses and latch circuits. For example, if the data width is 8 bits, data stored in 32 memory cells will be read out simultaneously. If 32 memory cells are read simultaneously, 32 latch circuits are required to latch the read data, and 3
Two buses are needed. Such an increase in the number of latch circuits and buses is caused not only when the data width is large, but also when the data width is large.
This also occurs when the number of times of continuously outputting data (burst length) increases. For example, in the case where an 8-bit continuous access (burst length = 8) is performed instead of the 4-bit continuous access (burst length = 4) shown in the conventional example, assuming that the data width is 8 bits as described above, Since 64 memory cells are accessed simultaneously, 64 latch circuits and 64 buses are required.

To avoid such an increase in the number of latch circuits and buses, there is continuous access by 2-bit prefetch. The 2-bit prefetch is different from the above-described nibble mode in which all data to be output is read at a time, and the data to be output continuously is read in 2 bits at a time, and the output is performed while the data is being output. The data to be read is read. In other words,
As in the nibble mode, by selecting one CSL, not all data to be output is read out, but one CSL is selected and data for two consecutive outputs is read out and latched. While the data is being output, another CSL is selected in order to read out data to be output further continuously thereafter. In this method, even if the burst length is increased, only two outputs are read at the same time, so there is no need to increase the number of buses. However, in 2-bit prefetch, one CS
Since two consecutively output data are simultaneously read by L, if one of the simultaneously read data is output, the next output is always determined by the other data. Problem. This will be described with reference to FIG. In the figure, (a)-
(H) is an address, and CSL0 to CSL3 are (a) and (b), (c) and (d), (e) and (f), respectively.
This is a CSL for simultaneously selecting (g) and (h). In the figure, an address different only in the least significant bit is selected by one CSL. However, regardless of the combination of addresses selected at the same time, two addresses selected by one CSL are hardened. Naturally, it is fixed. As an example, consider a case where the burst length is 4 and the input CPU address is (c). In this case, CSL1 is selected first, and the data stored in (c) and (d) is read. Latch these data,
While outputting in the order of (c) → (d), the selection of CSL is changed from CSL1 to CSL0, and the data stored in (a) and (b) is read. Therefore, the output order is (c) → (d) → (a) → (b)
It becomes the order of. That is, starting from the data of the CPU address (c), data of addresses different only in the lower two bits are sequentially output, responding to the request of the cache memory. However, if the CPU address is (d), CSL1 is selected first. Of the addresses (c) and (d) accessed,
When (d) is output, the next option is to output (c).
It can be seen that it is impossible to output the addresses in the order of (d) → (a) → (b) → (c). Therefore, when it is required to continuously output data in the order of addresses (hereinafter, referred to as sequential access), this method cannot satisfy this requirement. Similarly, it is possible to output in the order of (e) → (f) → (g) → (h), but not in the order of (f) → (g) → (h) → (e). In other words, it can be seen that sequential access is possible from an even address, but not from an odd address.

As a memory selection circuit capable of sequential access from an odd address, there is a 2-bit prefetch pipeline system using two sets of 2-bit prefetch circuits. In this system, a latch circuit and a bus are further provided. An increase is inevitable.

As described above, in the related art, a memory selection circuit that performs continuous access by n-bit prefetch has performed a single access using one CSL, and thus requires many latch circuits and buses. Sequential access from odd addresses was impossible or difficult.

Accordingly, it is an object of the present invention to provide a memory selection circuit which enables continuous access from an odd address and does not require many latch circuits and buses.

[0013]

According to the present invention, there is provided a memory selection circuit for performing continuous access by n-bit prefetch, wherein a means for multiple-selecting a column selection line for selecting only an address of data to be output at one output timing is provided. Have. Such means receives an input address and information indicating the length (burst length) of data to be continuously output, and sets a column selection line corresponding to the input address and a column corresponding to the address of data to be output continuously. The selection line and the multiple selection lines are simultaneously selected.

As described above, in the present invention, one column selection line selects only the address of data to be output at one output timing, and simultaneously selects a plurality of addresses.
Even in the case of using a prefetch method in which data to be output is read a plurality of times, the combination of data to be prefetched is arbitrary, and the output order is not limited.
For this reason, sequential access from odd addresses can be performed without increasing the number of buses and latch circuits.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS First, an embodiment of the present invention will be described in detail.

This embodiment is different from the conventional 2-bit prefetch in which data of two outputs is read by one column select line (hereinafter referred to as CSL), and CSL which reads only data of one output. Are simultaneously selected to read data for two outputs.

The memory selection circuit according to the present embodiment includes a prefetch predecoder for decoding lower bits of a column address. The prefetch predecoder receives information indicating a burst length in addition to an address, and based on the received information, indicates a burst length. , And a predecode address at which two bits of the address become active level is supplied to the column decoder. The column decoder receives the predecode address and the remaining upper bits of the column address, and simultaneously selects two CSLs connected to the memory cells to be read at the same time, thereby reading data for two outputs at the same time.

Next, this embodiment will be described in detail with reference to the drawings. FIG. 1 is a diagram showing an entire memory according to the present embodiment, in which 100 is an overall view of a 16MDRAM, which is integrated on a semiconductor chip. 101 is 16M
It is a memory cell array of bits and has 4096 word lines and 512 CSLs (CSL0 to CSL511). Therefore, since one address corresponds to eight memory cells, one input / output (data width) is eight bits. 1
02 is an address buffer, and 103 is a row decoder. Reference numeral 104 denotes a prefetch predecoder which is the main part of the present embodiment, and controls the multiple selection of CSL. 105 is a pre-decoder and 106 is a column decoder. 107
Is a clock generator which receives a signal as shown in the figure from the outside and generates a signal for controlling the timing of the prefetch predecoder 104 and the like. 108 is an I / O switch, 109 is an output buffer, 110 is an input buffer, and 111 is a latch circuit.

Next, the 16 MDRAM according to the present embodiment will be described.
Will be described taking the case of reading as an example. First,
The address buffer 102 that has received the input addresses A0 to A11 given from outside,
A11 is supplied to the row decoder 103, and A0 to A8 are supplied to the prefetch predecoder 104 as column addresses. The row decoder 103 receiving the row addresses A0 to A11 decodes this, and selects one of the 4096 word lines. On the other hand, column addresses A0 to A8
Then, the prefetch predecoder 104 receives the upper bits A3 to A8 of the column address and outputs them to the predecoder 1
05, and predecodes the lower bits A0 to A2 and outputs them as predecode addresses Y0 to Y7. The operation of the prefetch predecoder 104 is based on signals LOAD0, 1,
2 and the COUNT signal.
Here, the LOAD0 signal is a signal generated in response to the rising of the clock from the outside, and the LOAD1
The AD2 signal is a signal generated in synchronization with the clock when the externally applied signal is a read or write command, and the COUNT signal is LOAD1, LOA.
This signal is generated according to the burst length after the generation of the D2 signal. It should be noted that other control signals are also output from the clock generator 107, but a description thereof will be omitted. The decode output from the predecoder 105 and the predecode address from the prefetch predecoder 104 are both supplied to the column decoder 106, and the CSL is selected by these. The prefetch predecoder 104 is further supplied with burst signals B4 and B8 indicating a burst length. Based on these burst signals, the prefetch predecoder 104 corresponds to an address corresponding to an input address and data to be output next. A predecode address in which two bits of the address are at the active level is output so that the address corresponding to the address is selected. Therefore, the column decoder 106 receives the decode output from the predecoder 105 and the predecode address in which the two bits of the address are at the active level, and
Two SLs are selected at the same time. This CSL is
As described above, each is connected to eight memory cells.
The memory cells are read simultaneously. All of these 16-bit data read simultaneously are latched by the latch circuit 111 and supplied to the I / O switch 108. Further, from the prefetch predecoder 104, which CSL of the two CSLs selected at the same time is
Signal CI0 indicating whether to read a cell connected to SL
Is supplied to the I / O switch 108, and based on this signal, among the data latched by the latch circuit 111, the data of the memory cell connected to one CSL is output to the output buffer 109, and the signal CI0 is output. The data of the memory cell connected to the other CSL is output to the output buffer 109 in response to the change. And the output buffer 1
09 is output to the outside as outputs D0 to D7 of the 16MDRAM. While these data are being output, the clock generator 10
7, the predecode address changes, the selection of CSL is changed, and other data is accessed.

The above is the outline of the operation of the 16MDRAM shown in the present embodiment. Hereinafter, the configuration and operation of each unit will be described in more detail. FIG. 2 is a diagram showing the inside of the prefetch predecoder 104 in detail. The prefetch predecoder 104 includes an address latch generator block 280 and a predecode block 290.
The address latch generator block 280 receives the column addresses A0 to A8,
The internal address signals ADD0 to ADD2 are generated from 0 to A2. In the figure, Ai indicates A3 to A8. The predecode block 290 has an internal address signal A
Receiving DD0 to ADD2, predecode addresses Y0 to Y0
Y7 is generated. In the figure, 201 to 203 are latch circuits, 204 is a selection signal generation circuit, and 205 is a counter. The latch circuit 201 has nine latch circuits 2
The latch circuits 201-3 receive the column addresses A3 to A8, respectively.
201 to i are collectively shown as 201-i. Similarly, the latch circuit 202 includes nine latch circuits 202-0 to 202-8.
2-3 to 202-8 are collectively shown as 202-i. The latch circuit 203 includes seven latch circuits 203-0, 203-3 to 202-8. The latch circuit 201 takes in data when the LOAD0 signal goes to the active level, the latch circuit 202 takes in data when the LOAD1 signal goes to the active level, and the latch circuit 203 and the selection signal generating circuit 204 respectively
When the AD2 signal becomes an active level, data is fetched. The counter 205 is a 3-bit counter, and is composed of flip-flop circuits 205-0, 205-1 and 205-2 to which weights of 1, 2, and 4 are assigned, respectively. That is, the counter 205 is a counter that counts from 0 to 7. The counter 205 has L
OAD2 signal and COUNT signal are supplied,
Data is fetched in response to the active level of the LOAD2 signal, and counted up in response to the COUNT signal. The carry from the flip-flop circuit 205-1 to the flip-flop circuit 205-2 is performed via the AND gate 270 as shown in the figure. The burst signal B8 input to the AND gate is:
This signal indicates that the burst length is 8 when it is 1 (high level). Therefore, the flip-flop circuit 2
The carry from 05-1 to the flip-flop circuit 205-2 is not performed except when the burst length is 8. Similarly, the burst signal B4 is a signal indicating that the burst length is 4 when it is 1 (high level). Burst signal B4
When both B8 and B8 are 0 (low level), it indicates that the burst length is 2. Also, COUN
The T signal is also supplied to the selection signal generation circuit 204,
The data stored in the selection signal generation circuit 204 is COU
It is inverted every time the NT signal becomes active.

FIG. 3 is a diagram showing the column decoder 106 in detail. The column decoder 106 includes 64 switch circuits 106-0 to 106-63, and each of the switch circuits is commonly supplied with predecode addresses Y0 to Y7 from the prefetch predecoder 104. Further, a decode output from the predecoder 105 is supplied to each switch circuit. FIG. 4 is a diagram showing a part of the predecoder 105. In FIG.
x and Ay are column addresses A3 and A4, or column addresses A5 and A6, or column addresses A7 and A8. These three sets of column addresses are decoded, and Ax0Ay0,
Ax1Ay0, Ax0Ay1, and Ax1Ay1. This is input to the column decoder 106 as shown in the figure, and the column decoder 106 receives the decode signal and the predecode addresses Y0 to Y7 and selects the CSL. FIG. 5 shows one of the switch circuits 106-k constituting the column decoder 106 in detail. FIG.
, 501 to 504 are P channel M, respectively.
The OS transistors 505 to 508 are N-channel MOS transistors. Switch circuit 106
The decode outputs A3XA4X, A5XA6X, and A7XA8X from the predecoder 105 input to −k are transistors 503 and 505, 502 and 506, respectively.
Input to the gates of 501 and 507. Therefore, these decoded outputs A3XA4X, A5XA6X,
When all of A7XA8X become 1 (high level), the potential of the contact 510 becomes VSS (0). As a result, the input predecode addresses Y0 to Y7 are output as CSL via the buffer circuits 521-0 to 521-7. Decode output A3XA4X, A5XA6
If at least one of X and A7XA8X is 0 (low level), the buffer circuits 521 to 528 always output 0.

FIG. 6 is a diagram showing a connection relationship between CSL and bit lines. As shown, each CSL is connected to only one bit line pair. Further, it can be seen that reference numerals 601 and 602 denote I / O bus pairs, and adjacent bit lines are connected to different I / O bus pairs.
That is, since the present embodiment performs 2-bit prefetch, two bit line pairs selected at the same time are those connected to the I / O bus pair 601 and those connected to the I / O bus pair 602. Thus, two bit line pairs are selected, and 2-bit data is latched in latch circuit 111. In this embodiment, since the data width is 8 and there are eight input / output terminals, only two pairs of I / O buses are shown in the figure, but this is for one input / output terminal. I / O
There will be 16 bus pairs. Therefore, in the figure, C
SL is connected to only one bit line pair, but a bit line connected to another input / output terminal, that is, another I / O
Since they are also connected to the bit lines connected to the bus pair, they are actually connected to eight pairs of bit lines.

Returning to FIG. 2, the input addresses A2, A2,
Taking the case where A1 and A0 are respectively 0, 1, 1 (3) as an example, the read operation will be described more specifically for cases where the burst length is 2, 4, and 8, respectively.

First, the case where the burst length is 2 is shown in FIG.
This will be described with reference to FIG. 7 which is a timing chart. First, the column addresses A0 to A8 supplied from the address buffer 102 are latched by the latch circuit 201 in response to the generation of the LOAD0 signal. Therefore, the latch circuits 201-2, 201-1 and 201-0 have A2, A
1, A0 are latched, and 0, 1, 1 are stored, respectively. Thereafter, when a READ command is supplied from the outside through a combination of the CS, RAS, CAS, and WE signals, L
OAD1 and OAD2 signals are generated, and column addresses A0 to A8 are generated.
Are latched by the latch circuit 202, and the column addresses A3 to A8 are also latched by the latch circuit 203. However, the column addresses A2, A1, and A0 are flip-flop circuits 205-2 and 205-1 and a selection signal generation circuit, respectively. 204. Also, burst signals B4, B
Since both 8 are 0, 0 is output from the OR gate 240, and 0 is always latched in the latch circuit 203-0 even if the LOAD2 signal occurs. Since the flip-flop circuit 205-0 is connected to VSS, the initial value of such a circuit is always 0. Note that the latch circuit 202 is used for detecting the end when performing interleave access. Since the above-described data is latched in each latch circuit and flip-flop circuit, 1 is output from the AND gate 212. Accordingly, the predecode addresses are 1 for Y2 and Y3, and 0 for the others. As described above, two corresponding CSLs are selected by the predecode address thus generated, 16-bit data is latched by the latch circuit 111, and the I / O switch 10
8 is supplied. At this time, since the selection signal CI0 is 1, among the 16-bit data supplied to the I / O switch 108, 8-bit data read by the CSL corresponding to the Y3 of the predecode address is selected, and the output buffer is selected. It is output to 109. When the burst length is 2, the COUNT signal is generated once from the clock generator 107 in synchronization with the clock after the generation of the LOAD1 signal. The generation of the COUNT signal inverts the selection signal CI0 to 0, selects the 8-bit data read by the CSL corresponding to the predecode address Y2, and outputs it to the output buffer 109. At this time, the counter 205 also counts one, and 0, 1, and 1 are stored in the flip-flop circuits 205-2, 205-1, and 205-0, respectively, but this does not change the predecode address. Of course. After all, in this case, in terms of the column address, ... 0
11 (3),... 010 (2).

Next, a case where the burst length is 4 will be described with reference to FIG. 2 and a timing chart of FIG. In this case, since the burst signal B4 is 1 and B8 is 0,
The AND gate 260 stores the address A0 in the latch circuit 203.
Supply to −0. The burst length is 2 in the selection signal generation circuit 204 and other latch circuits and flip-flop circuits.
The same data as in the case of is stored. As a result, 0 and 1 are output from the exclusive OR gates 250 and 251, respectively, and 1 is output from the AND gate 213. At this time, since the burst signal B4 is 1, 1 is also output from the AND gate 220, and the generated predecode address is 1 for Y0 and Y3 and 0 for the others. The data read by the predecode address is latched by the latch circuit 111, and the CSL corresponding to the predecode address Y3 is first read by the selection signal CI0 supplied to the I / O switch 108 as described above. The bit data is selected and output to the output buffer 109. When the burst length is 4, the COUNT signal is generated three times from the clock generator 107 in synchronization with the clock after the generation of the LOAD1 and LOAD2 signals. First, when the COUNT signal is 1
Times, the selection signal CI0 is inverted to 0,
The output is switched to 8-bit data read by the CSL corresponding to the predecode address Y0. Subsequently, when another COUNT signal is generated, 0, 0, 0 are stored in the flip-flop circuits 205-2, 205-1, 205-0, respectively, and 1 is output from the AND gate 211. Therefore, Y1 and Y2 are 1
Then, the others are changed to the predecode address of 0.
At this time, since the selection signal CI0 is inverted twice and returns to 1, the I / O switch 108 selects and outputs 8-bit data read by the CSL corresponding to the predecode address Y1. When the last COUNT signal is generated, the selection signal CI0 is similarly inverted to 0,
The output is switched to 8-bit data read by the CSL corresponding to the predecode address Y2. To summarize, first, the predecode addresses are Y0 and Y3
Is 1 and Y1 and Y3, Y1 and Y2, and Y1 and Y2 become 1 each time the COUNT signal is generated. In response, first, 8-bit data read by the CSL corresponding to Y3 is output to the output buffer 109,
Every time the UNT signal is generated, 8-bit data read by the CSL corresponding to Y0, Y1, Y2 is output. Therefore, in terms of column address,
011 (3), ... 000 (0), ... 001 (1),
... 010 (2) are output in this order.

The same applies to the case where the burst length is 8, but when the burst length is 8, the counter 205 counts up to 205-2 as described above. Accordingly, when the burst signal B4 is 0 and B8 is 1, 1 is output from the AND gate 213 and AND gate 223, and the predecode addresses are Y3 and Y3.
4 becomes 1. When the burst length is 8, the LOA
After the generation of the D1 and D2 signals, the clock generator 10
7, the COUNT signal is generated seven times in synchronization with the clock. Therefore, every time the COUNT signal is generated, the predecode addresses which become 1 are Y3 and Y4, Y5 and Y6, Y5 and Y6, Y0 and Y7, Y0 and Y7, Y1 and Y2, Y1 and Y
It changes to 2. Accordingly, the output buffer 109 is
First, 8-bit data read by the CSL corresponding to Y3 is output, and every time a COUNT signal is generated, Y4,
CSL corresponding to Y5, Y6, Y7, Y0, Y1, Y2
Will be output. Therefore, in terms of column address,... 011
(3), 100 (4), 101 (5), 1
10 (6), ... 111 (7), ... 000 (0), ...
.. 001 (1),... 010 (2).

In summary, FIG. 10 shows that the burst length is 2,
For each of cases 4 and 8, it shows what predecode address the prefetch predecoder 104 outputs to the input addresses A0 to A2. In addition,
The predecode address shown in FIG. 10 indicates an address that becomes 1.

As described above, when the prefetch predecoder 104 is used, when the burst length is 2, the output starts from the data corresponding to the input address, and the data corresponding to the address in which only the least significant bit of the address differs. If the burst length is 4, data is output starting from the data corresponding to the input address, and data corresponding to an address in which only the lower 2 bits of the address are different is output in the address order and output. In the case of 8, the output starts from the data corresponding to the input address, and the data corresponding to the address in which only the lower 3 bits of the address are different is output in the address order. For this reason, even if any address is input, it is possible to output data of a continuous address in which only the lower bits of the address are different from each other in order of address, so that it is possible to cope with a case where a sequential access is requested. Moreover, since 16 bits are read out by selecting two CSLs,
Even if the burst length is long, there is no need to increase the bus as in the nibble mode.

The address latch generator block 280 shown in FIG. 2 may be a circuit shown in FIG. An address latch generator block 281 shown in FIG. 11 is obtained by replacing the AND gate 260 in the address latch generator block 280 shown in FIG. 2 with switches 1101 and 1102, and the function thereof is the same as that of the address latch generator block 280 shown in FIG. Is the same as However, the parts related to the upper bits A3 to A8 of the column address are completely the same as those in FIG. In the figure, burst signals SL4 and 8 are signals that become 1 (high level) when the burst length is 4 or 8, and burst signals SL4 and 8B are burst signals SL4 and 8B.
8 is an inverted signal. Also, switches 1101 and 1
Reference numeral 102 denotes a switch that closes (turns on) when these burst signals are “1”. Therefore, when the burst length is 2, the VSS potential (0) is latched in the latch circuit 206, and when the burst length is 4 or 8, the data of A0 is latched.

Further, the address latch generator block 280 may be a circuit shown in FIG. However, as in FIG. 11, the parts related to the upper bits A3 to A8 of the column address are completely the same as in FIG. The address latch generator block 282 shown in FIG. 12 can perform interleave access in addition to the above-described continuous access (sequential access). Switching between the sequential access and the interleave access is performed by the access control signals FSL and FIL. When the access control signal FSL is 1 (high level) and the access control signal FIL is 0 (low level), the sequential access is performed. When the signal FSL is 0
When the access control signal FIL is 1 (high level) at (low level), interleave access is performed. In the figure, switches 1201 to 1210 have the same functions as switches 1101 and 1102 shown in FIG. First, in the case of sequential access, in this case, switches 1201, 1203, 1208, and 1209 are closed, and switches 1202, 1204, and
Since 1207 and 1210 are open, they perform the same operation as the address latch generator block 281 shown in FIG. On the other hand, in the case of the interleaved access, the switches 1201, 1203, 1208 and 1209 are open, and the switches 1202, 1204,
Since 1207 and 1210 are closed, the initial values of the flip-flop circuits 205-0, 205-1 and 205-2 always become the VSS potential (0). Further, the data of the latch circuits 205-1 and 205-2 are directly used as the internal address signal ADD as in the case of the sequential access.
1 and ADD2, respectively.
Exclusive OR outputs with the data latched in 02-1 and 202-2 are generated by exclusive OR gates 1220 and 1230, and are used as internal address signals ADD1 and ADD2, respectively.

On the other hand, predecode block 290 shown in FIG. 2 may be a circuit shown in FIG. The predecode block 291 shown in FIG. 13 is provided with an internal address signal ADD0 from the address latch generator block.
2 is different from the application decoding block 290 shown in FIG.
223 is replaced by switches 1321 to 1324. Although the function is the same as that of the predecode block 290 shown in FIG. 2, the predecode block 291 shown in FIG. 13 has a simple circuit configuration, so that higher integration is possible.

Next, another embodiment of the present invention will be described.

In the present embodiment, 4-bit prefetch is performed by multiple selection of CSL instead of 2-bit prefetch. The prefetch predecoder 1400 shown in FIG. 14 is replaced with the prefetch predecoder 1400 shown in FIG.
It is realized by using it in place of 04. The prefetch predecoder 1400 is the same as the prefetch predecoder 104 in that the prefetch predecoder 104 receives the lower three bits of the column address and generates the predecode addresses Y0 to Y7. 1 for 2 (high level)
, While the prefetch predecoder 1
400 differs in that four of the predecode addresses Y0 to Y7 are set to 1. As described above, since the prefetch predecoder 1400 sets four of the predecode addresses Y0 to Y7 to 1, the column decoder 106 receives this and selects four out of the 512 CSLs. This means that the memory cells are read simultaneously.

Next, the prefetch predecoder 1400
Will be described. Prefetch predecoder 1400
Is an address latch generator block 1440 that receives the lower three bits A0 to A2 of the column address to generate internal address signals ADD0 to ADD2, and a predecode that receives the internal address signals ADD0 to ADD2 and generates predecode addresses Y0 to Y7. Block 145
It consists of 0. In the figure, 201 to 203, 1412
And 1414 are latch circuits, 1410 and 1411
Is a selection signal generation circuit, and 1441-1444 are switches, and their functions are the same as those of the same circuit described above.
The selection signal generation circuit 1411 is a 2-bit counter composed of flip-flop circuits 1411-0 and 1411-1, stores data in response to the LOAD2 signal, and counts up in response to the COUNT signal. However, the flip-flop circuits 1411-0 to 1
The carry to 411-1 is performed via the AND gate 1460. The burst signal B8 is 1 (high level) indicating that the burst length is 8, and the burst signal B4, 2 is 1 (high level) indicating that the burst length is 4 or 2. The signals, burst signals B4 and B8, are 1 (high level) signals indicating that the burst length is 4 or 8.

Also in this embodiment, the addresses A2, A1, A0 input to the prefetch predecoder 1400 are used.
Are respectively 0, 1, 1 (3) as an example, and the case where the burst length is 2, 4, and 8 will be specifically described.

First, the case where the burst length is 2 will be described. First, the column addresses A0 to A8 supplied from the address buffer 102 are latched by the latch circuit 201 in response to the generation of the LOAD0 signal. Therefore,
The latch circuits 201-2, 201-1 and 201-0 include:
A2, A1, and A0 are latched, and 0, 1, and 1 are stored, respectively. When the burst signal B4,2 is 1 and the switch 14
42 and 1443 are closed, the latch circuit 141 is closed.
VSS potential (0) is input to both 2 and 1413, and 1 is output from AND gate 1420. Therefore, the predecode address is 1 for Y0, Y1, Y2 and Y3. According to the predecode address generated in this manner, four corresponding CSLs are selected, and 32-bit data is latched by the latch circuit 111, and the I / O is performed.
The signal is supplied to the switch 108. On the other hand, the counter 1411
Flip-flop circuits 1411-0 and 1
411-1 has column addresses A0 and A1 respectively.
Are stored, the selection signals CI0 and CI1 are both 1. These selection signals are both supplied to the I / O switch 108, and the I / O switch 108 receives the selection signal and selects 8-bit data read out by the CSL corresponding to the predecode address Y3 from the 32-bit data. Output to the output buffer 109. Thereafter, the flip-flop 141 is generated by the COUNT signal generated once.
Since the data stored in 1-0 is inverted, the selection signal C
I0 is also inverted to 0 (low level). However, since the burst signals B4 and B8 are 0 and the AND gate 1460 outputs 0, the carry to the flip-flop circuit 1411-1 is not carried out, and the selection signal CI1 is not inverted. Thus, the generation of the COUNT signal causes the selection signals CI1 and CI0 to be 1 and 0, respectively, and the 8-bit data read by the CSL corresponding to the predecode address Y2 is selected and output to the output buffer 109. After all, in this case, in terms of the column address,... 01
1 (3),... 010 (2).

The case where the burst length is 4 is the same as the case where the burst length is 2. That is, AND gate 1
420 outputs 1 and the predecode address is Y
0, Y1, Y2 and Y3 become 1. Although the selection signals CI1 and CI0 are both 1, when the burst length is 4, the carry to the flip-flop circuit 1411-1 is carried out. The value of the existing counter 1411 is 00, 0
Accordingly, the I / O switch 108 starts outputting to the output buffer 109 from the 8-bit data read out by the CSL corresponding to the predecode address Y3, and selects Y0, Y1. , Y2. Therefore, in terms of column addresses,... 011 (3),... 000 (0),.
1 (1),... 010 (2).

On the other hand, when the burst length is 8, the switches 1441 and 1444 are closed, so that the latch circuit 14
Column addresses A0 and A1 are stored in 12 and 1413, respectively. Therefore, in this case, 1 is input to both latch circuits 1412 and 1413, and 1 is output from AND gate 1423. As a result, the predecode addresses of Y3, Y4, Y5, and Y6 become 1, and the selection signals CI1, CI0 of 11, 0 as described above.
I / O switch 10 according to the change of 0, 01, 10
Reference numeral 8 indicates that the output to the output buffer 109 is started from 8-bit data read by the CSL corresponding to the predecode address Y3, and the selection is changed in the order of Y4, Y5, and Y6. When the burst length is 8, the COUNT signal is generated seven times. Therefore, after the data corresponding to the predecode address Y6 is output, the counter 1410 responds to the COUNT signal generated continuously, and the counter 1410 responds to the COUNT signal. Carry up to two. Therefore, the internal address signal ADD2 is inverted to 1 and, as a result, 1 is output from the AND gate 1427.
As for the predecode address, Y7, Y0, Y1, and Y2 become 1. Thereafter, similarly, in response to a change in the selection signals CI1 and CI0, the corresponding predecode address is Y
7, Y0, Y1, and Y2 are output from the I / O switch 108 to the output buffer 109 in this order. Therefore, in terms of column addresses,... 011 (3),.
100 (4),... 101 (5),... 110 (6),
... 111 (7), ... 000 (0), ... 001
(1),... 010 (2).

As described above, similarly to the prefetch predecoder 104, the prefetch predecoder 1400 can output data of continuous addresses in which only lower bits of the address are different according to the burst length in the order of addresses. In addition, since 32 bits are read out by selecting four CSLs, even if the burst length is long, it is not necessary to increase the number of buses as in the nibble mode. It can be performed.

[0040]

As described above, the present invention has means for multiplexing the column selection lines for selecting only the addresses of data to be simultaneously output in one output, so that the input address is an even address. The data to be output first and the data to be output next can be prefetched regardless of whether the address is an odd address. Further, even if the burst length becomes long, it is not necessary to increase the number of buses and the like, so that an increase in chip size can be prevented. Therefore, the memory selection circuit according to the present invention is most suitable for a memory such as a synchronous DRAM which performs continuous access.

[0041]

[Brief description of the drawings]

FIG. 1 is an overall view of a memory according to an embodiment of the present invention.

FIG. 2 is a circuit diagram showing a prefetch predecoder 104.

FIG. 3 is a diagram showing an entire column decoder 106;

FIG. 4 is a diagram showing a part of a predecoder 105.

FIG. 5 is a diagram showing a part of a column decoder 106.

FIG. 6 is a diagram showing a connection relationship between a column selection line and a bit line.

FIG. 7 is a diagram showing a read timing according to an embodiment of the present invention (burst length = 2).

FIG. 8 is another diagram (burst length = 4) showing read timing according to an embodiment of the present invention.

FIG. 9 is still another diagram (burst length = 8) showing read timing according to one embodiment of the present invention.

FIG. 10 is a diagram showing a predecode address generated by a prefetch predecoder 104.

FIG. 11 is a diagram showing another example of the address latch generator block.

FIG. 12 is a diagram showing still another example of the address latch generator block.

FIG. 13 is a diagram showing another example of a predecode block.

FIG. 14 is a diagram showing a prefetch predecoder 1400 according to another embodiment of the present invention.

FIG. 15 is a diagram showing a nibble mode according to a conventional example.

FIG. 16 is a diagram showing addresses accessed in 2-bit prefetch according to the conventional example.

[Explanation of symbols]

100 Overview of 16M DRAM 101 16 Mbit memory cell 102 Address buffer 103 Row decoder 104, 1400 Prefetch predecoder 105 Predecoder 106 Column decoder 107 Clock generator 108 I / O switch 109 Output buffer 110 Input buffer 111, 201 to 203, 1412, 1413 Latch circuit 204, 1411 Selection signal generation circuit 205, 1410 Counter 210-217, 220-223, 260, 270, 1
300 to 1307, 1420 to 1427 AND gate 230 to 237, 240, 1333 to 1337 OR gate 250, 251, 1220, 1230 Exclusive OR gate 280 to 282 Address latch generator block 290, 291 Predecode block 1101, 1102, 1201 to 1210, 1321-
1324, 1441 to 1444 Switches A0 to A11 Input address ADD0 to ADD2 Internal address signal Y0 to Y7 Predecode address B4, B8, SL4, 8, SL4, 8B, B4, 2, B
4,8 Burst signal LOAD0, LOAD1, LOAD2, COUNT
Timing signal FSL, FIL Access control signal

Claims (5)

(57) [Claims] 1. A memory selection circuit for receiving an input address composed of a plurality of bits and activating a predetermined column selection line among a plurality of column selection lines based on the input address, wherein a least significant bit of the input address is latched. Latch means for receiving, a counter means for receiving a plurality of bits different from the least significant bit of the input address and counting it up in response to a count signal, at least the least significant bit and the counter latched by the latch means Activating means for simultaneously activating a plurality of column selection lines among the plurality of column selection lines in response to a count value of the means. Wherein at least the plurality of column selection lines in which the least significant bit and the count-up latched by the latch means are activated simultaneously in response to the count value of said counter means prior to, the 2. The memory selection circuit according to claim 1, further comprising a first column selection line corresponding to an input address and a second column selection line corresponding to an address different from the input address. 3. The memory selection circuit according to claim 2, wherein said different address is an address next to said input address. Wherein at least said latch means the least significant bit and said plurality of column select lines to be activated simultaneously in response to the count value of said counter means after the count-up is latched by, the The method according to claim 1, further comprising a third column selection line different from any of the first and second column selection lines and a fourth column selection line different from any of the first to third column selection lines. Item 3. The memory selection circuit according to item 2 or 3. 5. An output of said latch means and a burst signal.
Circuit that latches and outputs a signal generated based on
The output of the latch circuit and the counter
In response to the count value of the stage, the activation means
Activating the column selection lines simultaneously
Item 2. The memory selection circuit according to Item 1 .