EP0698312A1 - Tile based architecture for fpga - Google PatentsTile based architecture for fpga
- Publication number
- EP0698312A1 EP0698312A1 EP19950909504 EP95909504A EP0698312A1 EP 0698312 A1 EP0698312 A1 EP 0698312A1 EP 19950909504 EP19950909504 EP 19950909504 EP 95909504 A EP95909504 A EP 95909504A EP 0698312 A1 EP0698312 A1 EP 0698312A1
- European Patent Office
- Prior art keywords
- Prior art date
- 230000002457 bidirectional Effects 0 description 1
- 230000015572 biosynthetic process Effects 0 abstract 1
- 239000000872 buffers Substances 0 claims description 22
- 239000000460 chlorine Substances 0 description 7
- 230000001721 combination Effects 0 description 2
- 230000000295 complement Effects 0 claims description 6
- 230000001276 controlling effects Effects 0 claims description 11
- 239000011162 core materials Substances 0 claims description 69
- 230000000875 corresponding Effects 0 description 2
- 230000001808 coupling Effects 0 claims 7
- 238000010168 coupling process Methods 0 claims 7
- 238000005859 coupling reaction Methods 0 claims 7
- 230000001186 cumulative Effects 0 description 1
- 238000009826 distribution Methods 0 description 1
- 230000035611 feeding Effects 0 description 1
- 238000005755 formation Methods 0 abstract 1
- 230000003993 interaction Effects 0 description 1
- 238000005304 joining Methods 0 description 1
- 239000011133 lead Substances 0 description 3
- 239000011159 matrix materials Substances 0 abstract claims description 120
- 230000015654 memory Effects 0 claims description 40
- 238000009740 moulding (composite fabrication) Methods 0 claims description 7
- 230000000737 periodic Effects 0 description 1
- 229920002795 polyhydroxyoctanoate Polymers 0 description 2
- 230000003405 preventing Effects 0 claims 2
- 239000000047 products Substances 0 description 1
- 230000002829 reduced Effects 0 description 1
- 230000001603 reducing Effects 0 description 5
- 238000006722 reduction reaction Methods 0 description 2
- 230000004044 response Effects 0 description 1
- 239000010703 silicon Substances 0 description 3
- 239000007787 solids Substances 0 description 1
- 238000003860 storage Methods 0 description 2
- 239000000758 substrates Substances 0 description 1
- 238000007514 turning Methods 0 description 2
- H03—BASIC ELECTRONIC CIRCUITRY
- H03K—PULSE TECHNIQUE
- H03K19/00—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits
- H03K19/02—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits using specified components
- H03K19/173—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits using specified components using elementary logic circuits as components
- H03K19/177—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits using specified components using elementary logic circuits as components arranged in matrix form
- H03K19/1778—Structural details for adapting physical parameters
- H03K19/17796—Structural details for adapting physical parameters for physical disposition of blocks
- H03—BASIC ELECTRONIC CIRCUITRY
- H03K—PULSE TECHNIQUE
- H03K19/00—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits
- H03K19/02—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits using specified components
- H03K19/173—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits using specified components using elementary logic circuits as components
- H03K19/177—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits using specified components using elementary logic circuits as components arranged in matrix form
- H03K19/17704—Logic circuits, i.e. having at least two inputs acting on one output; Inverting circuits using specified components using elementary logic circuits as components arranged in matrix form the logic functions being realised by the interconnection of rows and columns
TILE BASED ARCHITECTURE FOR FPGA
FIELD OF THE INVENTION _^ The invention relates to programmable logic devices formed in integrated circuits and more particularly to an architecture of a programmable logic device in which logic blocks are provided in a repeating pattern.
BACKGROUND OF THE INVENTION Field programmable gate arrays (FPGAs) are well known in the art. An FPGA comprises an array of configurable logic blocks (CLBs) which are programmably interconnected to each other to provide a logic function desired by a user. U.S. Patent 4,870,302, reissued aε U.S. Patent Re.34,363, and incorporated herein by reference, describes a well known FPGA architecture. Other publications, such as U.S. Patent 4,758,745, U.S. Patent 5,243,238, and published application WO 93/05577, also incorporated herein by reference, describe other FPGA architectures. The Xilinx 1993 Data Book entitled "The Programmable Logic Data Book", available from Xilinx, Inc., 2100 Logic Drive, San Jose, California 95124, also incorporated herein by reference, describes several products which implement a number of FPGA architectures. An FPGA is considered to be a general purpose device, i.e. being capable of performing any one of a plurality of functions, and is programmed by an end user to perform a selected function. Because of this design flexibility, a general purpose FPGA includes a significant number of wiring lines and transistors which remain unused in most applications. Moreover, FPGAs include overhead devices which facilitate programing of the FPGA to do the specified function. These overhead devices undesirably add area to the FPGA chip. To compensate for this overhead, it is commercially important to reduce the cost of the FPGA. One way to reduce the cost is to make the FPGA less general purpose, that is, to eliminate some configuration options which are less commonly used. However, this reduction in configuration options reduces the value of the FPGA to customers, who may not be able to predict which options will be needed. Therefore, a need arises to eliminate area while maximizing configuration options.
SUMMARY OF THE INVENTION In accordance with the present invention, a field programmable gate array (FPGA) architecture includes repeatable tiles. Each tile comprises a programmable routing matrix and a configurable logic block matrix. The configurable logic block matrix is programmably connectable to the programmable routing matrix, as well as to the configurable logic block matrices in adjacent tiles. The programmable routing matrix is programmably connectable to the programmable routing matrices adjacent to the tile, as well as to long lines which extend across the tile. Thus, each tile provides a combination of logic, connection to nearby tiles, and connection to a general routing structure. A plurality of these tiles are joined together to form an array of tiles which make up the functional portion of an FPGA chip. With this architecture, devices of different sizes are produced by simply joining together different numbers of tiles, thereby eliminating an expensive and time consuming design effort. Moreover, in accordance with the present invention, the programmable routing matrix and configurable logic block matrix minimize the number of programmable interconnection points (PIPs), thereby reducing expensive chip area and maximizing density of the entire chip. In further accordance with the present invention, proper positioning of the PIPs ensures the necessary routing flexibility, thereby maximizing functionality of the FPGA. A tile architecture has a set of signal lines exiting the tile at the boundaries. Thus, for example, signal lines exiting at the right of one tile connect with signal lines exiting at the left of another tile. In one embodiment, adjacent tiles are identical, forming a repeating pattern. In another embodiment, adjacent tiles are not identical but have signal lines at least most of which match at the tile boundaries. Thus, a chip can be formed as an array of modular units which match at their boundaries, and additional flexibility of designing tiles for use in a plurality of chip designs iε easily available.
BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 shows an FPGA chip which includes components' according to the present invention. Fig. 2A shows a single core tile which populateε a majority of the FPGA chip illustrated in Fig. 1. Fig. 2B shows four adjacent core tiles of the type illustrated in Fig. 2A. Fig. 3A illustrates a configurable logic block matrix which is part of the tile of Fig. 2A. Fig. 3B illustrates a multiplexer structure which implements all PIPs which connect the output lines of a configurable logic block to one output line. Fig. 3C shows one embodiment of a multiplexer structure which drives a configurable logic block input line. Fig. 4C illustrates the configurable logic block in the matrix of Fig. 3A. Fig. 4B illustrates tri-state buffer block 302 of Fig. 3A. Fig. 4C illustrates the output enable block 309 of Fig. 3A. Fig. 4D showε a look up table embodiment of the F, G, H and J function generatorε of Fig. 4A. Fig. 4E εhowε another look up table embodiment of the F, G, H and J function generatorε of Fig. 4A. Fig. 4F shows one Karnaugh map for the look up table function generator of Fig. 4D or 4E. Fig. 4G showε one of the 216 logic functionε which can be implemented by the look up table function generator of Fig. 4D or 4E. Figs. 5A-5C illustrate application of the configurable logic block of Fig. 4A to form a carry chain, a cascadable decode circuit, and two 5-input combinational functions, respectively. Fig. 6 illustrateε the programmable routing matrix of Fig. 2A. Fig. 7A illustrateε an example of the connectivity achieved by a programmable routing matrix of the invention such as shown in Fig. 6. Fig. 7B illustrates an example of the connectivity achieved by the combination of the programmable routing matrix of Fig. 6 and the tile structure of Fig. 2A or 2B. Fig. 8 illustrateε connectionε from global signal pads near corners of a chip to global signal lines which extend near four edges of the chip and connect to global lines which drive core tiles. Fig. 9 illustrates long line splitters which are provided on long lines in one embodiment of the invention. Figs. 10A-10D illustrate, respectively, left, top, right, and bottom edge tiles according to one embodiment of the invention. Figε. 11A-11D illustrate upper left, upper right, lower right, and lower left corner tiles for the same embodiment. Fig. 12 illustrated a logic diagram for one embodiment of the oscillator structure used in Fig. 11B.
DETAILED DESCRIPTION OF THE DRAWINGS The following drawing conventionε are used throughout the figures. A εmall solid black dot at the intersectionε of two lines indicates a permanent electrical connection between the crosεing lines. An open circle enclosing an intersection between two lines indicates a programmable connection between the lines (for example, a pass transistor which is turned on to make the connection) . Open circles represent bidirectional signal flow between the two lines. An open triangle at an intersection of two lines indicates a programmable connection with signal flow going onto the line pointed to by the apex of the triangle. (The signal is of course then present on the full length of the line. Thus, a triangle pointing in the opposite direction would have the same signal flow because the triangle points to the same wire.) In accordance with one embodiment of the present invention, programmable connections are provided by using programmable interconnection points (PIPs), wherein each PIP includes at leaεt one transistor. A triangle on a line which is not intersected by another line indicates a buffer which produces signal flow in the direction indicated by the apex of the triangle. Except for global lines ENOUT and ENLL (illustrated in Fig. 3A) , a line which ends within the tile or matrix structure (i.e. does not extend to the border of a tile or matrix) is physically terminated within the tile. A line which extends to the border of the tile or matrix connects to a line on the next tile, which it contacts when two tiles are abutted together. Note that some lines which extend to an edge of a tile and thus into an adjacent tile change names at the tile boundary. Lines in the configurable logic block matrix and the programmable routing matrix are given the εame reference numeral to indicate theεe lineε are phyεically connected to each other. Fig. 1 shows an FPGA chip 100 according to the present invention.' In the center portion of chip 100 are a plurality of identical core tiles 101, which are interconnected by conductive lines (described in detail below) . Along the four edges of chip 100 are west, north, east, and south edge tiles 103, 104, 105, 106, respectively. In the four corners of the chip are four corner tiles 113, 114, 115, and 116. Chip 100 includes pads, i.e. pads P1-P56, for connecting edge tiles 103, 104, 105, 106, and corner tiles 113-116 to external pins of a package (which holds chip 100) . Note that each edge tile is further connected to a core tile 101. As shown in Fig. 1, edge tiles are connected to different numbers of pads P, typically from zero to four pads (explained in detail in reference to Figε. lOa-lOd) . Fig. 1 alεo illuεtrateε high voltage source pads VCC and low voltage source pads GND. Power and ground connections (not εhown) are provided in a conventional manner throughout chip 100. Fig. 2A shows a core tile 101. Core tile 101 includes a programmable routing matrix 201 and a configurable logic block matrix 202. Programmable routing matrix 201 iε described in detail in reference to Fig. 6, whereas configurable logic block matrix 201 iε described in detail in reference to Fig. 3A. Extending to the west from programmable routing matrix 201 are twelve lines with suffixes 0 through 11. These include single length west lines W1-W5, W7-W11, and double length west lines DW0 and DW6 (described in detail below) . Extending to the north from programmable routing matrix 201 are single length north lines N1-N5, N7-N11 and double length north lines DN0 and DN6. Extending to the east are single length east lines E1-E5 and E7-E11 and double length east lines DEO and DE6. Extending to the south are single length south lines S1-S5 and S7-S11 and double length south lines DSO and DS6. Extending east to west acrosε tile 101 are double length horizontal lineε DH0 and DH6. Extending north to εouth across tile 101 are double length vertical lines DV0 and DV6. Fig. 2B shows four adjacent core tiles 101a, 101b, 101c and lOld having a configuration identical to tile 101 illustrated in Fig. 2A. For clarity in Fig. 2B, most lines are not labeled. As mentioned previously, lines extending to the edges of tile 101 connect to lines in adjacent tiles. For example, single length west line Wlb in tile 101b extending to the west from programmable routing matrix 201b connects to single length east line Ela in adjacent tile 101a. Double length horizontal line DH6a of tile 101a is coupled to double length west line DW6b of tile 101b, and is further coupled to a double length eaεt line DE6 of a tile not εhown in Fig. 2B but which iε located directly west of tile 101a (hence the terminology "double length"). Line QOc extending east from CLB matrix 202c in tile 101c connectε to line QWOd extending weεt from CLB matrix 202d in tile lOld. Fig. 2B also illustrates that horizontal global lines GH0 and GH1 and vertical global lineε GVO and GVl extend continuouεly from one tile 101 to the next. Theεe global lines may be connected to a common line at the edge of the tile εo that a signal on a global line such aε GHO extends through all tiles. As εhown in Fig. 2B, vertical global lines GVO and GVl and horizontal global lines GHO and GH1 are coupled to both programmable routing matrix 201 and configurable logic block matrix 202. Returning to Fig. 2A, configurable logic block (CLB) matrix 202 is connected to the CLB matrix in the west tile (not shown) by output lines Q0-Q3 and input lines QW0-QW3, to the CLB matrix in the north tile (not shown) by output lines Q0-Q3 and input lines QN0-QN3, to the CLB matrix in the east by output lines Q0-Q3 and input lines QE0-QE3, and to the CLB matrix in the south tile (not shown) by output lines Q0-Q3 and input lines QS0-QS3. Note that output lines Q0-Q3 carry the εame signalε from CLB matrix 202 to adjacent tiles in four directions and thus have the same nameε. Carry-in line CIN and carry-out line COUT, which extend vertically in tile 101, connect to carry-out and carry-in lines, respectively, in adjacent tiles to form a fast carry path for arithmetic functions, as diεcuεsed in detail in U.S. Patent No. 5,349,250, "LOGIC STRUCTURE AND CIRCUIT FOR FAST CARRY", which is incorporated herein by reference.
Configurable Loσic Block Matrix 202 Fig. 3A illustrateε configurable logic block (CLB) matrix 202 of Fig. 2a. CLB matrix 202 includes a CLB 301, a tristate buffer block 302, an input interconnect εtructure 303, a CLB output interconnect εtructure 304, a feedback interconnect εtructure 305, a general input interconnect εtructure 306, a register control interconnect structure 307, an output interconnect structure 308, and an output enable block 309.
Sparse Pioulation Programmable connections are provided by using programmable interconnection points (PIPs) , wherein each PIP includes at least one transistor. As is well known in the art, each transistor occupieε valuable space on the chip substrate. Thus, in accordance with the present invention and referring to Fig. 3A, a majority of the horizontal and vertical lines in input interconnect structure 303, feedback interconnect structure 305, general input interconnect structure 306, and register control interconnect structure 307 are not programmably connectable. In other words, these structureε are εparsely populated with PIPs, or are sparεely "pipulated" . Sparεe pipulation minimizeε chip area uεed by PIPs, thereby maximizing density of the entire chip. In further accordance with the present invention, proper positioning of the PIPs significantly increases routing flexibility, thereby effectively compensating for the reduced number of PIPs in the interconnect structures. For example, referring to input interconnect structure 303, PIPs are positioned to allow connection from each output line Q0-Q3 from CLB output interconnect structure 304 to one of the function generators F, G, H, or J of an adjacent tile in each of the four compasε directionε. In this embodiment, general input interconnect structure 306 provides four to six PIPs for each CLB input line (J0-J3, JB, H0-H3, HB, G0-G3, GB, F0-F3 and FB) to CLB 301. Feedback interconnect structure 305 provides direct connections from two of output lines Q0-Q3 to one of the function generator input terminals in CLB 301. As shown in Fig. 3A, 24 PIPs in output interconnect structure 308 connect output lines Q0-Q7 to tile interconnect lines M0-M23. In thiε manner, signals on tile interconnect lines M0-M23 are selectively transferred between CLB 301 and programmable routing matrix 201 (via CLB output interconnect structure 304, general input interconnect structure 306, and output interconnect structure 308) . In this embodiment, less than one intersection in eight is provided with a PIP, thereby minimizing silicon area. Yet, connectivity from any output line to any input line is ensured by the PIPs provided.
Configurable Loσic Block 301 A configurable logic block (CLB) 301 is illustrated in Fig. 4A. In this embodiment, CLB 301 includes four function generators F, G, H, and J, wherein each function generator comprises a 16-bit look up table that generates an output signal determined by the four input signalε provided to the function generator and the valueε stored in the look up table. Thus, function generator F generates an output signal determined by the input signalε provided on lineε F0-F3, function generator G generateε an output εignal determined by the signals provided on CLB input lines G0-G3, function generator H generates an output signal determined by the signals provided on CLB input lines H0-H3, and function generator J generates an output signal determined by the signals provided on CLB input lines J0-J3.
Look UP Table Operation of the look up table function generators will be described in connection with Figε. 4D-4G. These figures were firεt diεcussed by Freeman in U.S. Patent 4,870,302 now reisεued as U.S. Patent Re 34,363, incorporated herein by reference. Fig. 4D illuεtrateε a look up table, in this embodiment a 16-bit RAM, which provides an output signal in response to any one of sixteen possible combinations of four input signals. Specifically, input signals A and B control the X decoder to select any one of the four columns in the 16-bit RAM. In a similar manner, input signals C and D control the Y decoder to select any one of the four rows in the 16-bit RAM. The 16-bit RAM provides an output signal representative of the bit at the intersection of the selected row and the selected column. There are 16 such intersections and thus sixteen such bits. It logically follows that 16 bits provide 216 posεible combinationε. Thus, if a 4-input NOR gate is to be simulated by the 16 bit RAM, the Karnaugh map for the look up table would be as shown in Fig. 4F. In Fig. 4F, all bits are "0" except the bit at the intersection of the first row (representing A=0, B=0) and the first column (representing C=0, D=0) . If a logic "1" output signal iε deεired for A=l, B=0, C=0, D=0, then a logic "1" is stored at the intersection of the second row and the firεt column. If a logic "1" iε desired for A=0, B=0, C=0, and D=0 and also for A=l, B=0, C=0 and D=0, then a logic "1" is stored at each of the intersections of the firεt column with the first row and the second row. The logic circuit represented by this loading of the look up table is shown in Fig. 4G. Thus, the look up table of Fig. 4D represents an elegant and simple implementation of any one of 2i6 logic functions. Fig. 4E showε a regiεter configuration for yielding any one of εixteen select bits. Each of regiεterε 0-15 in the vertical column to the left labeled "16 Select Bitε", contains a selected signal, either a logic 1 or 0. By selecting the appropriate combination of signals A, B, C, and D and their complements, a particular bit stored in a particular one of the εixteen locations in the 16 Select Bits register is transmitted to the output lead OUT. Thus, for example, to tranεmit the bit in the "1" regiεter to the output lead, the signal A, B, C, D is applied to the leads so labeled. To transmit the εignal labeled "15" in the εixteenth location in the 16 Select Bits register to the output lead, the signal A, B, C, D is applied to the appropriate columns. Thus, this regiεter configuration also provides any one of 216 logic functions. Referring back to Fig. 4A, the memory bits in look up tables F, G, H and J are typically loaded during configuration of the chip, for example through a shift register, or alternatively by an addresεing means. In some embodiments, the memory bitε are also loaded during operation of the chip, thereby reconfiguring the chip on the fly. A reconfigurable memory structure is discuεεed in commonly aεεigned, U. S. Patent No. 5,343,406 invented by Freeman et al. and entitled "Diεtributed Memory Architecture for a Configurable Logic Array and Method for Using Distributed Memory", which is incorporated herein by reference. Function generators F, G, H, and J provide output signals on CLB output lines X, Y, Z, and V, respectively. These output εignalε from function generatorε F, G, H, and J control multiplexerε Cl, C2, C3, and C4, thereby providing a cumulative carry-out function COUT. Multiplexer Cl receiveε a carry-in εignal on line CIN and an input εignal on line FB, and generateε an output εignal on line CF. Multiplexer C2 receiveε the εignal on line CF and an input εignal on line GB, and generateε an output εignal on line CG. Multiplexers C3 and C4 are connected in the same manner aε Multiplexerε Cl and C2. Multiplexer C4 provideε an output signal on line COUT from CLB 301. For a detailed discuεεion of the implementation of arithmetic functionε, see commonly asεigned U.S. Patent No. 5,349,250 invented by Bernard E. New, entitled "LOGIC STRUCTURE AND CIRCUIT FOR FAST CARRY", which iε incorporated herein by reference. In addition to function generatorε F, G, H, and J, each CLB 301 includes four storage devices RX, RY, RZ, and RV. These storage devices RX, RY, RZ, and RV each comprise flip flops with master and slave stages and an output multiplexer which takes outputs from the master and slave stages as inputs. Thus devices RX, RY, RZ, and RV can be configured by the multiplexer to serve aε either flip flopε or as latches. Typically, periodic repowering of the carry signal is neceεεary. In this embodiment, to provide this repowering, a repowering buffer comprising inverters 1121 and 1122 is positioned every four multiplexers in the carry path, or once every CLB 301. In another embodiment, a repowering buffer is provided every two multiplexerε in the carry path, thus two repowering buffers are provided in every CLB 301. In this embodiment, CLB 301 includes five input lines per function generator. For example, referring to function generator F, CLB input lineε F0-F3 provide input εignalε to function generator F, and a fifth CLB input line FB provides a multiplexer control input signal. Function generators G, H, and J are configured in a similar manner. Three input lines CLK, CE, and RST provide clock, clock enable, and reεet εignalε, reεpectively, to regiεterε RX, RY, RZ, and RV. Aε shown in Fig. 4A, four groups of three output εignalε are provided from CLB 301, one group associated with each function generator. The three output signals include: «a direct, unregistered output signal from the function generator (provided on CLB output lines X, Y, Z, or V) , »an alternative, unregiεtered output εignal which may be derived from one of the CLB input εignals, a signal from the carry chain, or in two caεeε a εignal from a multiplexer which provideε an output εignal of a five- input function (provided on CLB output lineε XB, YB, ZB, or VB) , and »a registered, output signal which may be loaded by the function generator or by one of the sources of the alternative output signal (provided on CLB output lines XQ, YQ, ZQ, or VQ) . For example, CLB output line X receives a direct unregiεtered output signal from function generator F. CLB output line XB receives either the signal on CLB input line FB or the output signal of multiplexer SI (as determined by multiplexer Bl) , which in turn iε derived from either the carry-out εignal CF or the five-input function-generator output signal from multiplexer FG (see discussion of Fig. 5C below) . CLB output line XQ receives the regiεtered output signal from register RX, which derives its D input signal either directly from function generator F (the signal on output line X) or the alternative output signal on line XB as determined by multiplexer DI. Finally, output line K provideε a constant signal, which may be high or low, as selected by multiplexer PG. In the embodiment of Fig. 4A, multiplexers D1-D4 selectively provide either the output signals from function generators F, G, H, and J (the same signals on CLB output lines X-V) or the output εignalε from multiplexerε B1-B4 to registers RX-RV, reεpectively. If multiplexerε SI and S3 are set to forward the carry signalε of multiplexerε Cl and C3 , reεpectively, then multiplexerε B1-B4 εelect between the input signals on CLB input lines FB-JB, respectively, and the output signals of multiplexers C1-C4. Multiplexers C1-C4, in addition to being used for the carry function in an arithmetic operation, alεo generate wide AND and OR functions. To generate the AND function, a logic 0 is placed on line FB to program multiplexer Cl to generate an AND function of the F function generator output signal on CLB output line X and the carry-in signal on line CIN. Alternatively, to generate the OR function, a logic 1 is placed on CLB input line FB to program- multiplexer Cl to generate an OR function of the complement of the output signal on CLB output line X and the carry-in signal on line CIN. With a truth table architecture, the OR function is achieved by loading the inverse values into the truth table. The function of multiplexers C1-C4 and their interaction with the logic block are further discussed in application serial no. 08/116,659 [M-2565] incorporated by reference.
Example Applications of CLB 301 Figs. 5A-5C illustrate applications using CLB 301 (described in detail in reference to Fig. 4A) to form a carry chain, a cascadable decode circuit and 2 five-input functions, respectively. Theεe figures use heavy lines to illustrate lines of CLB 301 which are used for the particular selected function and thin dashed lines to indicate lines and elements not used for the particular function. In Fig. 5A, CLB 301 is configured to compute a half sum H3H2H1H0 (where H3, H2, Hi, and HO are the four bits of a four-bit half-sum) and the carry bitε C3C2C1C0 of two numbers A3A2A1A0 and B3B2B1B0. Another CLB (not shown), preferably poεitioned in the tile to the right or left of the one shown, will be used to complete the sum. Operands A3 and B3 are placed on any two of CLB input lines J0-J3. Operands A2 and B2 are placed on any two of CLB input lines H0-H3. Al and BI are placed on any two of CLB input lines G0-G3. A0 and B0 are placed on any two of CLB input lines F0-F3. Unused lineε are either held high or held low. Each of function generatorε F, G, H, and J is loaded with the truth table of the XOR function (which is the half sum of its input signals) . The truth table takes into account the values applied to unused input lineε. If there are lower order bits than those applied to function generator F, the carry-out of those bits is placed on carry in line CIN. Multiplexers Cl, C2, C3, and C4 are controlled by the output signals of function generators F, G, H and J, respectively. Specifically, if the function generator output signal is a logic 1 (εignalε A and B are not equal) , the carry-in value iε forwarded to the carry-out of that bit, and if the function generator output signal is a logic 0 (signals A and B are equal), the value of signal A or signal B is forwarded to the carry-out of that bit. Multiplexers B1-B4, SI and S3 are controlled to forward the carry-out of each bit to the "B" CLB output line (i.e. CLB output lines XB, YB, ZB, and VB) of that bit. The function generator output signal for each bit (on CLB output lines X, Y, Z, and V) is provided as the half sum output for that bit. In another application shown in Fig. 5B, CLB 301 is configured to operate as a cascadable decoder. A 16-bit addresε repreεented by εignals A0-A15, iε placed on CLB input lines F0-F3, G0-G3, and J0-J3. CLB input lines FB, GB, HB, and JB are grounded. The 16 bits of each of function generatorε F, G, H, and J include a εingle logic 1 to reflect a portion of a predetermined addreεε. A logic 1 εignal iε placed on carry in line CIN. If all four function generatorε F, G, H, and J output their respective logic Is (i.e. indicating an address "match"), then multiplexers C1-C4 all forward a logic 1 and produce a logic 1 signal on carry out line COUT. In yet another application εhown in Fig. 5C, CLB 301 is configured to generate two functions of five input signalε each. Function generatorε F and G generate a firεt function of five input εignals on CLB output line XB and function generators H and J generate a second function of five input signals on CLB output line ZB. For the first function, four input signalε A0-A3 are provided on the CLB input lines to both function generators F and G and the fifth input signal A4 is provided to line FB. Input signal A4 causes multiplexer FG to select the output signal of function generator F or function generator G. In this embodiment, multiplexer SI is programmed by its memory cell to select the output signal of multiplexer FG, and multiplexer Bl iε programmed by its memory cell to select the output signal of multiplexer SI. Thuε, the five-input function output εignal from function generators F and G is provided on CLB output line XB. In a similar manner, the function of the five input signals- B0-B4 provided to function generators H and J is generated on CLB output line ZB. Loading the appropriate truth tables into the two asεociated function generators F and G produces the desired function of five input signals. Specifically, in one embodiment, a 32-bit look up table is stored in function generators F and G (i.e. two 16-bit look up tableε) . Thus, a large number of functions are alternatively provided by loading different values into the memory cellε which form the truth tableε of the function generatorε and control multiplexerε FG and HJ.
Triεtate Buffer 302 Fig 4B illustrates a schematic drawing of tri-state buffer block 302 (Fig. 3A) which includes tristate buffers B4- B7. Note that the line names are identical to those referenced in Fig. 3A. Output εignals from AND gates A4-A7 control tristate buffers B4-B7, respectively. If AND gate A5, for example, provides a logic 0 output signal, buffer B5 is enabled and provideε a buffered output signal on line TQ5 which matches its correεponding input signal on line Q5. On the other hand, if AND gate A5 provides a logic 1 output signal, buffer B5 is disabled and provides a high impedance at the output terminal. The output signalε provided by AND gates A4-A7 are determined either globally by the output signal from OR gate ORl or individually by memory cells MM4-MM7, respectively. If memory cells MM4-MM7 store logic O'ε, then the output signals of AND gates A4-A7 will also be logic O's regardlesε of the εignal from OR gate ORl. OR gate ORl provideε a high output εignal if' the ENLL εignal iε low or if the signal on line TS is high. Referring back to Fig. 3A, the signal on tristate line TS is programmably selected from any of tile interconnect lines M16-M23. The ENLL signal is a global εignal provided to all bufferε 302 in all tileε 101. The ENLL εignal iε held low during configuration and as other signals are being enabled after configuration in order to prevent contention which could result if various TS lineε which are to connect .input εignalε to the εame long line are εwitching unpredictably during configuration. If bufferε B4-B7 are to be used during operation as repowering buffers (always enabled) for placing a signal onto a long line, memory cellε MM4-MM7 are loaded with low valueε during configuration. Thiε meanε that during configuration, AND gateε A4-A7 will enable buffers B4-B7. However, no contention occurs because the input signalε Q4-Q7 which drive εignalε TQ4-TQ7 onto long lineε all carry a common εignal during configuration, aε will now be diεcuεεed in connection with Fig. 4C.
Output Enable Block 309 The bufferε in output enable block 309 are disabled during configuration of the device so that lines driven by these buffers will not experience contention. Fig. 4C illustrates the structure of block 309. Each buffer in output enable block 309 compriseε a two-input AND gate. One input of each AND gate is driven by a global enable εignal ENOUT. The other input iε provided by a line Q0'-Q7' which iε in turn provided by output εignalε from CLB 301 (Fig. 3A) . During configuration, unexpected lineε may be connected to theεe lines Q0-Q7. Therefore, to prevent contention, the ENOUT signal is held low during configuration so that all output εignalε on lineε Q0-Q7 are low and unexpected connection of other lines does not produce contention because all signalε have a low value.
Neighbor Input Matrix 303 Referring back to Fig. 3A, in accordance with this embodiment of the present invention, adjacent CLBs 301 are not connected via direct connectionε, only via PIPε. For example, input signals are selectively provided to CLB 301 from input interconnect structure 303. Thus, each input line QS0-QS3 is connectable to one of the CLB input lines of one function generator. In this embodiment, line QS0 is connectable to CLB input line Fl of function generator F, line QSl is connectable to CLB input line Gl of function generator G, line QS2 is connectable to CLB input line HI of function generator H, and line QS3 is connectable to CLB input line Jl of function generator J. Because each function generator F, G, H or J is configurable to provide any function based on its input signals, a particular signal can be provided to any input terminal of a function generator and the look up table of that function generator loaded accordingly. Thus, it is not important which input signal is available to which function generator input terminal. A signal on input line QW0 drives both CLB input lineε F0 and FB. Similarly, a signal on input line QWl driveε CLB input lines GO and, GB, a signal on input line QW2 drives CLB input lines HO and HB, and a signal on input line QW3 drives CLB input lines JO and JB. Each signal on input lines QE0, QE1, QE2, and QE3 also drives two CLB input lines. Specifically, a signal on input line QE0 drives CLB input lines Fl and FB, a signal on input line QE1 drives lines Gl and GB, a signal on input line QE2 drives lines HI and HB, and a signal on input line QE3 driveε lineε Jl and JB. Signals on input lines QN0-QN3 and QS0-QS3 each drive only one CLB input line. Specifically, a signal on input line QN0 drives CLB input line F0, a signal on input line QN1 driveε CLB input line GO, a εignal on line QN2 drives CLB input line HO, and a signal on line QN3 drives CLB input line JO. A signal on input line QSO drives CLB input line Fl, a signal on input line QSl drives CLB input line Gl, a signal on input line QS2 drives CLB input line HI, and a signal on input line QS3 drives CLB input line Jl. This embodiment is particularly desirable for horizontal flow of many signalε becauεe each input line QE0-QE3 and QW0-QW3 iε programmably connected to two CLB input lineε. Other embodiments of the present invention, having a different number and positioning of programmable connections, are optimized for a different signal flows.
Output Matrix 304 CLB 301 provideε output εignalε on CLB output lineε X, XQ', XB, Y, YQ, YB, Z, ZQ, ZB, V, VQ, and VB. Note that CLB 301 alεo deter ineε whether it provideε the signal on carry out line COUT or whether the signal on carry in line CIN is transferred to the next CLB in the tile above. PIPs on CLB output lines X, XQ, XB, Y, YQ, YB, Z, ZQ, ZB, V, VQ, VB, and K are selectively programmed to drive any number of output lines Q0-Q7 through a CLB interconnect structure 304. Note that CLB interconnect structure 304 is fully pipulated (i.e., any of the 13 output εignalε of CLB 301, excluding the εignal on carry out line COUT, can drive any of output lineε Q0-Q7) . Note that interconnect structure 304 also bufferε itε output εignals for driving further lines. Full pipulation of interconnect εtructure 304 requires 108 (13 x 8) PIPs. In contrast, structures 303, 305, 306, and 307 in combination use 200 PIPs, even though they are sparεely pipulated. Flexibility of CLB 301 to acceεε a particular input εignal from tile interconnect lineε M0-M23 iε ensured by: »fully pipulating CLB output interconnect εtructure 304 so that any CLB output signal can be provided to any of tile interconnect lineε M0-M23; opipulating programmable routing matrix 201 εo that each line M0-M23 is connected to at least one line M0-M23 in each adjacent routing matrix 201 (εee discusεion of Fig. 6 below) ; «pipulating CLB matrix 202 so that each output line of one CLB can be connected to at least one input line of each adjacent CLB; and •forming function generators F, G, H, and J as look up tables, thereby allowing all input signalε to each look .up table to be interchangeable. <»Moreover, except for five-input functionε, function generatorε F, G, H, J are alεo interchangeable. Thus, in accordance with the present invention, the above- described sparsely pipulated structureε 303, 305, 306 and 307 significantly reduces chip area while maximizing flexibility. Signals on output lines Q0-Q3 drive the input lines of CLBs in neighboring tileε. For example, by placing two core tileε 101 of Fig. 2A εide by side, as in εhown in Fig. 2B, output line Q0 on the left edge of core tile 101b connectε to input line QE0 on the right edge of tile 101a. Other lineε are correεpondingly connected. Thuε, referring to Figε. 2A, 2B, and 3 in combination, CLB output line X (Fig. 3A) of CLB 301 in CLB matrix 202c (εee Fig. 2B) is programmably connected to output line QOc, which extends east (as well as other directions) from CLB matrix 202c in core tile 101c, which in turn iε connected to input line QWOd of CLB matrix 202d in core tile lOld. PIPε are provided (aε diεcuεεed above) for connecting input line QWO to CLB input lineε F0 and FB of CLB 301. Thuε, in this manner, a path is establiεhed from the output lineε of CLB 301 in CLB matrix 202c to the input lines of CLB 301 in CLB matrix 202d using only two PIPs, which in one embodiment includes two transistors. In another embodiment, shown in Fig. 3B, a PIP in CLB output interconnect structure 304 requires a signal on a CLB output line to propagate through two transiεtorε (note that signal K, a constant power or ground εignal, propagates through four tranεiεtors) . Fig. 3B illustrates a multiplexer structure 400 which implements all PIPs which connect the twelve CLB output lines (X, XQ, XB, Y, YQ, YB, Z, ZQ, ZB, V, VQ, VB) of CLB 301 and one power/ground output signal line K to output line Q0. Multiplexer structure 400 includes memory cells 31, 32, and 33 which control a first bank of twelve transiεtors 351 and select signal K if no transiεtor in bank 351 iε selected. A logic 1 stored in one of memory cells 31, 32, and 33 selects one signal from each group of three signalε in bank 351. If all memory cellε 31, 32, and 33 store a logic 0, then signal K is provided to node 30. In a second εtage, memory cellε 34 and 35 control AND gateε AND1-AND4 to select the output signal from one of output lines VQ, ZQ, YQ, and XQ and to provide the selected signal on output line Q0. If ' memory cells 31, 32, and 33 store a logic 0, thereby selecting signal K, then memory cells 34 and 35 must be programmed to provide the signal at node 30. Thus, thirteen PIPs are implemented using only 5 memory cellε and sixteen transistors, each path requiring only two transistors for all signals except the conεtant value K, which travelε a longer path. The signal on line K iε not harmed by having a longer εignal path since it is not a switching signal. A multiplexer structure 400, which εelectε one of thirteen output εignals of CLB 301 to drive a predetermined output line, is provided for each of output lines Q0-Q7. Note that although it is poεεible for none of the thirteen output signals to drive an output line Q0-Q7, multiplexer structure 400 cannot select more than one of the thirteen output signals. In this manner, contention on output lines Q0-Q7 is avoided. In another embodiment of multiplexer structure 400, thirteen memory cellε are provided, each memory cell controlling a single transiεtor. In thiε manner, each path requireε only one tranεiεtor, thereby increaεing signal speed. However, note that this embodiment increaεeε silicon area.
Feedback Interconnect Structure 305 Referring back to Fig. 3A, feedback interconnect structure 305 selectively connects output lines Q0-Q3 to CLB input lines F2, G2, H2, and J2 within configurable logic block matrix 202. Thus, in this embodiment, any output signal from CLB 301 can be fed back to selected CLB input lines of any function generator F, G, H and J in CLB 301. Feedback interconnect structure 305 provides a PIP pattern that supports a counter (a counter feeds back its own signal) or a shift register (a shift register requires its neighbor'ε εignal) . The above-deεcribed PIP pattern prevents contention between εignalε on CLB input lineε F2, G2, H2 and J2 and εignals on CLB input lineε F0, GO, HO, JO, Fl, Gl, Hi, and Jl which are provided on other input lines to CLB matrix 202, such aε input lineε QWO and QN3. Other embodiments of the present invention provide different combinations of PIPs in feedback interconnect structure 305. •
General Input Matrix 306 General input matrix 306 receiveε input εignalε on tile interconnect lines M0-M23 and includes PIPs for placing these input signals onto CLB input lines F0-F3, FB, G0-G3, GB, HO- H3, HB, J0-J3, and JB. Optionally, a PIP pattern allows a signal on any tile interconnect line M0-M23 in general input interconnect structure 306 to drive one input line of each function generator F, G, H, and J. Because function generator input signals are interchangeable, (Lookup table inputs are interchangeable.) no tile interconnect line M0-M23 need be coupled to more that one input line of a function generator. In this embodiment of general input interconnect structure 306, PIPs are provided so that each CLB input line FB, GB, HB, and JB is driven by a signal on one of six tile interconnect lineε M0-M23. Aε another criterion in this embodiment, no CLB input line includes more than eight PIPs. Thus, referring to Fig. 3C, a multiplexer structure 401, using only three memory cells 36, 37 and 38, selects one of eight posεible εignalε to control a first bank of transiεtorε 361. Specifically, memory cell 38 εelectε one each of the paired εignalε on input lineε QWO or QN0, M15 or M14, M9 or M8, and M7 or M6. Memory cellε 36 and 37 provide εignals to the input terminals of AND gates AND5-AND8, which in turn control a second bank of transiεtors 362 to select a single εignal to place on CLB input line F0. In this embodiment of the present invention, the pattern of PIPs also provides a function of five inputs (discussed above in connection with Fig. 5C) . For example, a signal on tile interconnect line Ml8 or Ml9 driveε input line FB, a εignal on tile interconnect line M14 or M15 driveε lineε FO and GO, a εignal on tile interconnect line M12 or M13 driveε lines Fl and Gl, a signal on tile interconnect line M16 or Ml7 drives input lineε F2 and G2 and a εignal on tile interconnect line M20 or M21 driveε input lineε F3 and G3. In thiε configuration, five-input functionε are easily implemented with the PIP pattern provided. In further accordance with the present invention, and referring to Figs. 3A and 6, PIPs allow connection from long horizontal lines LH0-LH7 and long vertical lines LV0-LV7, as well as global (horizontal and vertical) lines GHO, GH1, GVO, and GVl to registerε RV, RZ, RY, and RX without going through function generators J, H, G, and F. Specifically, long horizontal lines LH0-LH7 and long vertical lines LVO and LV7 as well as global horizontal lines GHO, GHl and global vertical lines GVO, GVl are selectively coupled to tile interconnect lines M0-M23 (Fig. 6) . These tile interconnect lines, if coupled to CLB input lines FB, GB, HB and JB, bypass function generators F, G, H and J, respectively, and provide signals (via intermediate multiplexers) to registerε RX, RY, RZ, RV, reεpectively (Fig. 3A) . Note that global lines GHO, GHl, GVO, and GVl are also selectively coupled to registers RX, RY, RZ and RV via register control interconnect structure 307. Allowing all tile interconnect lineε M0-M23 to connect to one CLB input line FB, GB, HB or JB and providing connectionε from every long line to one tile interconnect line M0-M23 (diεcuεsed below in connection with Fig. 6) assureε that εignals on those long and global lines can drive the necessary registerε. In the preεent invention, thiε PIP pattern alεo allowε εignalε on all long lineε and global lines to drive input lines to function generatorε F, G, H and J via general input interconnect εtructure 306. Output Interconnect Matrix 308 In thiε embodiment, output lines Q4-Q7 also provide output signals to programmable interconnect matrix 201 (Fig. 2A) via tile interconnect lines M0-M11 or via lines TQ4-TQ7. Output lines Q0-Q3 also provide output signals to selected ones of tile interconnect lines M12-M23. In the embodiment shown in Fig. 3A, output interconnect structure 308 allows signals on each output line Q0-Q7 to drive up to three tile interconnect lines M0-M23. The full pipulation of CLB output interconnect structure 304 allows any utput line of CLB 301 to be connected to any tile interconnect line M0-M23. Note that general input interconnect structure 306 also provides selected feedback signals on output lines Q0-Q3 to CLB 301.
Register Control Interconnect Structure 307 Clock line CLK, clock enable line CE, reset line RST and tristate line TS may be driven by signalε provided on εelected tile interconnect lineε M0-M23 (from programmable routing matrix 201) . In addition, for a low εkew control, clock line CLK is driven directly by signalε on global horizontal lineε GHO and GHl or from global vertical lineε GVO and GVl.
No Contention In accordance with the preεent invention, if one PIP on a predetermined CLB input line iε programmed on, then no other PIP on that CLB input line εhould be turned on. For example, if the PIP at the interεection of input line QWO and CLB input line F0 iε programmed on (i.e. a εignal on input line QWO driveε CLB input line F0), then the PIPε on tile interconnect lines M6, M7, M8, M9, M14, Ml5, and input line QN0 remain turned off, thereby ensuring no contention on CLB input line F0. Typically, contention is avoided either by using a convenient decode method for selecting which PIP on a single input line is turned on or by uεing ruleε provided in the εoftware which programε the memory cellε to avoid turning on more than one PIP on an input line. In other embodiments, alternative input selection means are possible. For example, in one embodiment one memory cell is loaded to specify whether each PIP is turned on or not.
Programmable Routing Matrix 201 Fig. 6 illustrateε the programmable routing matrix 201 of Fig. 2a. Note that whereaε all PIPε in CLB matrix 202 are shown as triangles to indicate signal flow onto one line, in Fig. 6, most PIPs in programmable routing matrix 201 are shown aε open circleε to indicate εignal flow on both lineε. The exceptionε are PIPε which connect lineε TQ4 through TQ7 (output lineε from triεtate buffer block 302 of Fig. 3A) to long horizontal lineε LH0-LH7 and long vertical lineε LV0-LV7, and PIPε which place εignalε from global signal lines GHO, GHl, GVO, and GVl onto tile interconnect lines M0 through M3. Extending into programmable routing matrix 201 are global lines, long lines, double length lines, and εingle length lineε. Each of theεe lineε iε connectable to εelected tile interconnect lineε M0-M23. Programmable routing matrix 201 provideε connection to programmable routing matrices in adjacent tiles through single length lines extending in the four compasε directions, i.e. single length north lines Nl- Nil, single length east lines El-Ell, single length εouth lineε Sl-Sll, and εingle length weεt lineε Wl-Wll. Connection to programmable routing matriceε one tile away are provided by double length north lineε DN0 and DN6, double length eaεt lines DEO and DE6, double length south lines DSO and DS6, and double length west lineε DW0 and DW6 (see Fig. 2A) . Each long vertical line LV0-LV7 and long horizontal line LH0-LH7 which extends through the tile is connectable to one of tile interconnect lines M0-M23. The particular pattern of PIPs illustrated in Fig. 6 iε sparse, yet provides significant signal transferability. Specifically, programmable routing matrix 201, which in thiε embodiment includeε only 124 PIPs, is sparse relative to the approximately 4200 PIPε which could be provided to connect every line in Fig. 6 to every other line. However, the PIP pattern enεureε that any line iε connectable to any other line if enough intermediate PIPs are used. For example, as shown in Fig. 6, west line Wl is connectable to east line El by turning on two PIPs which connect tile interconnect line Ml to these two lines. In contrast, to make a connection between west line Wl and eaεt line E2 requireε 8 PIPε and 9 lines, i.e. connecting west line Wl to tile interconnect line Ml to eaεt line El to tile interconnect line M20 to weεt line W9 to tile interconnect line M9 to north line N9 to tile interconnect line M21, and finally to east line E2. Although a path of this length is typically undesirable, in some applications delay is unimportant. In those applications, the availability of such a path allows completion of a design. Easy paths requiring only two PIPs are available to connect lines Nl, SI, El, and Wl to tile interconnect line Ml; lines N2, S2, E2, and W2 to tile interconnect line M2 and so forth through tile interconnect line M5. Tile interconnect line M6 connects to double length lines DN6, DS6, DE6, and DW6. Tile interconnect lines M7 through Mil connect to correspondingly numbered single length lines extending north, south, eaεt and weεt. PIPs on tile interconnect lineε M12-M23 implement a pattern of cross connecting that facilitates signal transfer flexibility with minimal sacrifice of speed, and the sparεe pipulation achieveε valuable reduction of chip area. For example, tile interconnect line M12 connectε to double length north line DNO, to south line S3, to east line E5, and to west line Wl, whereas tile interconnect line M15 connects to north line N3, east line E8, double south line DS6, and west line W4. In this manner, the present invention provides a predetermined pattern to minimize the number of PIPs, thereby allowing any line to be connected to any other line. Thus, the present invention ensureε that a path is always provided, while minimizing silicon area.
Routing Matrix Model Each of tile interconnect lines M0-M23 is connectable to five or six other lines. Thus, as shown in Fig. 7A, each tile interconnect line M0-M23 is represented as a εtar with five or εix points. In this model, eight tile interconnect lines MO through M7 are programmably connectable to selected oneε of north lines N0-N3, eaεt lineε E0-E3, εouth lineε S0-S3 and weεt lineε W0-W3. Tile interconnect lineε MO through M3 are connectable to north, south, east and west lines of the same numerical suffix. Tile interconnect lineε M4 through M7 are connectable to εtaggered ones of the north, south, eaεt and west lines. Thus, tile interconnect lines M0-M3 provide a means for interconnecting north, east,- south and weεt lines of the same suffix, while tile interconnect lines M4-M7 provide an opportunity for cross-connecting lines from four compass directions. Also, tile interconnect lineε M0-M7 provide means for connecting programmable routing matrix 201 to configurable logic block matrix structure 202 (Fig. 3A) .
Connectivity Model for Routing Matrix and Logic Blocks Fig. 7B illuεtrateε the "εtar εtructure" of the present invention. In a star structure, each CLB 301 is asεociated with a particular εtar 201 (i.e the programmable routing matrix 201) from which radiate lineε connecting to other εtarε 201 and from there to other CLBε 301. In Fig. 7B, double length and εingle length lineε are illustrated. In other embodiments, lines of other lengths are provided in the εtar structure. Thus, the star structure of the present invention enεureε good connectivity between itε related CLBε and other partε of the device.
Global Interconnect Structure Fig. 8 illuεtrateε hard connections from global signal padε P113, P114, P115, and P116, which are poεitioned near the cornerε of chip 100 (Fig. 1) , to global signal lines GTL, GTR, GBR, and GBL, respectively, which are typically located near the four edges of chip 100. Each global εignal line is programmably connectable to a plurality of lines extending vertically or horizontally through each row or column of core tiles 101. For example, top left global signal line GTL is connectable to global vertical lines GVl-a through GVl-n, via PIPs PVl-a through PVl-n, respectively, i.e. one PIP for each column of core tiles 101. Top right global signal line GTR is connectable via PIPs PHO-a through PHO-m, respectively, i.e. one PIP for each row of core tileε 101 to global horizontal lines GHO-a. Bottom right global signal line GBR iε connectable to global vertical lineε GVO-a through GVO-n. Finally, bottom left global signal line GBL is connectable to global horizontal lines GHl-a through GHl-m. Note that the global vertical and horizontal lines with reference labels beginning with GV or GH are connectable to programmable routing matrices 201 and CLB matrices 202 in core tiles 101 through which the global lines pass, as discussed above in connection with Figs. 2A, 3, and 7. As also shown in Fig. 8, long lines LV0L, LV7L, LH0T, LH7T, LVOR, LV7R, LHOB, and LH7B which extend through the edge tiles (not shown in Fig. 8 for simplicity but shown in Figs. 10A through 10D) of chip 100 (Fig. 1) are also connectable to the global lines. Specifically, bottom right global εignal line GBR can be driven by εignalε on bottom horizontal long lineε LHOB and LH7B via PIPs PGBR0 and PGBR7, respectively. Bottom left global signal line GBL can be driven by signals on left vertical long lines LV0L and LV7L via bottom left buffer BBL via PIPε PGBL0 and PGBL7, reεpectively. Equivalent connectionε are provided for the top and right edgeε of the chip. Left, top, right, and bottom long lineε are connectable to each other through PIPε, such as PIP PBR7. Because long lineε LVOL, LV7L, etc. are driven by signals provided by any of the pads at the perimeter of the chip (through edge tiles 103-106 discussed below in connection with Figs. 10A-10D) , any pad can provide a global signal. Moreover, any of core tiles 101 can also provide a global signal through edge tiles 103- 106.
Optional Long Line Splitter Figs. 1 and 9 illustrate one embodiment of the preεent invention which includes long line splitters LLS which may be poεitioned partly through a line. Two columns of tiles are illustrated in Fig. 9, each column comprising a top edge tile 104, six core tiles 101, and a bottom edge tile 106. Long vertical lines LV0-LV7 traverse all core tiles 101, and in each of the two columns terminate in edge tileε 104 and 106. Long vertical lineε LV0-LV7 are alεo connectable to εelected oneε of tile interconnect lineε M0-M15 and lineε TQ0-TQ3 in edge tileε 104 and 106, aε will be discussed below in connection with Figs. 10A-10D. Furthermore, as diεcuεεed above in connection with Figε. 2A and 6, long lineε LV0-LV7 are connectable to εelected lineε in programmable routing matriceε 201. For clarity, horizontal long lineε LH0-LH7 are not illuεtrated in Fig. 9, but are illuεtrated in Figs. 2A and 6. In the embodiment shown in Fig. 9, vertical long lines LV0-LV7 in the three upper core tiles 101 are separated from the portions in the three lower core tiles 101 by long line splitterε LLS. An inεet illuεtrateε that a long εplitter LLS in one embodiment comprises an n-type tranεiεtor which iε turned off by providing a low voltage to a control gate CG, thereby separating the vertical long line into top and bottom segments. Long line splitterε LLS are typically uεed in large chip embodimentε to allow top and bottom long lineε to be separately driven in different portions of the chip. As shown in Fig. 1, horizontal long lines LH0-LH7 are also separated in the middle of chip 100 by long line εplitterε LLS. In other embodimentε, εeveral long line εplitters such as long line splitterε LLS and LLSA are provided along the εame long line, or long line splitters LLSB are provided between an end of a long lines in one edge tile and an end of a long line in an adjacent edge tile, thereby programmably connecting theεe long lineε.
Edge Tileε for Embodiment of Fig. 2A Figε. 10A-10D illuεtrate in greater detail the edge tiles shown in Fig. 2A. Specifically, Figs. 10A-10D show left edge tile 103, top edge tile 104, right edge tile 105, and bottom edge tile 106, respectively. Each edge tile in these embodimentε iε typically but not always connected to at least one of pads PV, PZ, PY or PX. In other embodiments described in detail below in reference to Fig. 1, at least one edge tile is not connected to any pad. In Fig. 10A, four pads, PV, PZ, PY, and PX are connected to edge tile 103 via input/output (I/O) devices IOBV, IOBZ, IOBY and IOBX, respectively. Each of I/O devices IOBV, IOBZ, IOBY and IOBX is connected to edge tile 103 by three line . For example, I/O device IOBV iε connected to edge tile 106 by an I/O input line IV, an I/O output line OV, and a tri-εtate line TSV. Note that the output signal provided to pad P42 by output line OV is controlled by a signal on I/O tri-state line TSV. Similar lines are provided for I/O devices IOBZ, IOBY and IOBX. A fully pipulated I/O input interconnect structure 1001 allows signalε on I/O input lines IV, IZ, IY, and IX to drive edge tile input lines QIN0-QIN3. Neighbor output interconnect structure 1004 allows signalε on output lines QE0-QE3 from a core tile 101 to be provided to pads PV, PZ, PY and PX. I/O output interconnect structure 1002 allows signals from the neighboring core tileε (in edge tile 103, provided by north lines 100-N7, south lines S0-S7, and east lineε E1-E5 and E7- Ell) aε well aε signals on long lines LH0-LH7 and LV0-LV7 and double length lines DHO, DH6, to be provided to the padε. Note that I/O output interconnect εtructure 1002 haε a εubεtantially complete pipulation, thereby allowing any εignal coming into left edge tile 103 from elεewhere in the chip interior to be placed on any of pads PV, PX, PY or PZ in spite of a sparse general interconnect structure 1006 between lines coming from other parts of the chip interior into or out of left edge tile 103 and a set of edge tile interconnect lines M0-M15. Intermediate interconnect structure 1003 allows signalε which come from one of tile interconnect lineε M0-M15 to be placed on one of edge tile input lines QIN0-QIN3, buffered onto a corresponding output line Q0 through Q3, and provided through tristate buffer block 302 to a corresponding line TQ0- TQ3. A signal can thence be provided to horizontal long lineε LH0-LH7 and vertical long lineε LV0-LV7. Thus, signalε on edge tile input lineε QIN0-QIN3 drive output lines Q0-Q3 directly and drive lines TQ0-TQ3 through tri-state buffer block 302. Feedback interconnect εtructure 1005 allowε signals on output lines Q0-Q3 to drive tile interconnect lines M0-Mi5 which are in turn selectively connected to north lines N0-N7, south lines S0-S7, east lines El-Ell, -double length lines DEO, DE6, DH0, and DH6 and to long lines LV0-LV7. In thiε manner edge tile 103 allowε connection to padε which in turn have external connectionε to chip 100, as well aε on an adjacent core tile 101 chip and to adjacent edge tileε (or an adjacent corner tile, explained in detail below) . Padε PV, PZ, PY, and PX repreεent padε P42, 41, 40 and P39, reεpectively, which are εhown in Fig. 1. Figε. 10B, 10C, and 10D εhow embodimentε of edge tileε 104, 105, and 106, respectively. Because these tiles are similar in structure, except for orientation, and have identical numerical references to that shown in Fig. 10A, the detail of the interface structureε in Figε. 10B, 10C, and 10D will not be diεcuεsed herein.
I/O Interface for Use With Optional Pad Fig. 10C illustrateε a combination of connected and unconnected padε, thereby illuεtrating the flexibility available at the maεk level. In thiε embodiment, one unconnected pad PZ and connected pads PV, PY, PX implement a configuration which iε repreεented in Fig. 1 by pads P6, P7 and P8 (connected to edge tile 105) . As shown in Fig. 1, each edge tile has a predetermined number of pads connected to it. For example, pad P17 is the only pad connected to its edge tile 106. Therefore, as shown in Fig. 10D, only one of padε PV, PZ, PY and PX (in thiε embodiment, pad PV) is connected to edge tile 106. Referring back to Fig. IOC, pad PZ and its input/output buffer structure IOBZ are eliminated, thereby reducing total chip size by reducing the total number of pads on the chip. Input line IZ and output line OZ are shorted together in a region which in one embodiment is outside tile 105. In this manner, all tiles 105 are identically laid out, regardless of how many pads PV, PZ, PY, or PX are provided. Referring back to Fig. 1, pads P6, P7 and P8 are connected to a single edge tile 105. In Fig. 10D, pad PY and related structureε IOBY and ESDY are not provided. Thuε, the embodiment of Fig. 10D represents pads P26 through P28 of Fig. 1. In other embodiments of the present invention, other pads are removed, up to and including removal of all four pads. For example, Fig. 1 includes certain edge tiles to which no pads have been connected (two of edge tiles 103, one of edge tiles 104, and one of edge tiles 105 have no pads at all connected to them) .
Corner Tiles Figs. 11A through 11D illustrate the four corner tiles 113, 114, 115, and 116, reεpectively, of chip 100 (Fig. 1) . Fig. 11A includeε a conventional boundary εcan block BSCAN compatible with IEEE 1149.1 deεcribed in detail in a Xilinx Application Note by Luiε Moraleε entitled, "Boundary Scan in XC4000 Deviceε" and available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124, which iε herein incorporated by reference in itε entirety. In Fig. 11A, top left corner tile 113 includes hard connections from εingle length eaεt lineε E0-E7 to εingle length εouth lineε S0-S7, reεpectively, and programmable connectionε from long horizontal lineε LH0-LH7 to long vertical lines LV0-LV7, respectively. Fig. 11A further εhowε one embodiment of an interconnect εtructure 1101 which provides the programmable connection of boundary scan block BSCAN to the above-described single length and long lines. Corner tile 113 also includes a programmable connection to an external pin P43 that provideε a global clock εignal SGCK1. Corner tile 114, illuεtrated in Fig. 11B, iε εimilar in configuration to corner tile 113 (Fig. 11A) . Specifically, tile 114 (Fig. 11B) includes hard connections for connecting single length west lineε W0-W7 to εingle length εouth lines S0-S7, respectively, and programmable connections for connecting long horizontal lines LH0-LH7 to long vertical lines LV0-LV7, reεpectively. In both Figε. 11A and 11B, long vertical line LVO connectε to long horizontal line LHO, but becauεe of the layout of tileε 113 and 114, the lineε are drawn in a different poεition on the page, and therefore corner tileε 113 and 114 have a different appearance in Figε. 11A and 11B. Corner tile 114 includes- a clock input pin Pl that provideε clock signal SGCK4. Corner tile 114 includes an interconnect structure 1102 which provides a programmable connection between a conventional oscillator/counter circuit DIV uεed for counting bitε during configuration of chip 100 and the above-described single length and long lineε. In one embodiment, circuit DIV iε used during chip operation to provide an on-chip oscillator or a counter-divider. Circuit DIV is typically configured to divide an internal oscillator signal or a user-provided signal. Corner tile 114 further includes a boundary scan update signal BSUPD, which is part of the standard boundary scan circuitry (most of the circuitry being located in tile 113) . In this embodiment, signal BSUPD is programmably placed on west lines W2 and W3 (and thus south lines S2 and S3) as well aε long horizontal lineε LH2 and LH3 (and thus long vertical lines LV2 and LV3) . Fig. 12 illustrateε one embodiment of a circuit which implementε oεcillator/counter circuit DIV of Fig. 11B. Two output tapε, OSCl and OSC2 are provided, which together can be configured to provide twelve frequencieε which are diviεionε of the original input frequency. An internal oscillator OSC provides an oscillator signal to NAND gate 1231. NAND gate 1231 iε enabled by a memory cell OSCRUN. When enabled, the output εignal from oscillator OSC is provided to multiplexer 1201. Memory cell 1202 determines whether multiplexer 1201 provides the output signal from internal oεcillator OSC or a εignal on one of εingle length west lines W0-W3 (equal to a signal on single length south lines S0-S3, respectively, see Fig. 11B) , or a signal on one of long horizontal lines LH0-LH3 (equal to a signal on long vertical lineε LV0-LV3) . Multiplexer 1201 provideε an output εignal which iε then available to be divided by flip flopε 1214 through 1220. Multiplexerε 1225 and 1226 provide a choice of divide factors on the data input terminals of flip flops 1227 and 1228 respectively. The outputs of theεe flip flops are provided as signalε on taps OSCl and OSC2. Flip flops 1227 and 1228 are clocked from the original input signal and serve to reduce the skew of the output signalε from multiplexers 1225 and 1226. Multiplexer 1225, under control of memory cells OSC1A and OSC1B, provides a switching signal which can be the input signal from multiplexer 1201 divided by 4, 16, 64, or 256. Depending upon the setting in memory cell 1203, multiplexer 1204 can forward the original clock εignal output from multiplexer 1201 or can provide a divided εignal (the original frequency divided by 512) which iε output from flip flop 1213. If multiplexer 1204 is set to provide the output signal of multiplexer 1201, then the original clock signal is alternatively provided by multiplexer 1226 as divided by 2, 8, 32, or 128. If multiplexer 1204 is εet to provide a divided εignal from flip flop 1213, multiplexer 1226 will provide an output εignal which haε the frequency of the original input εignal on multiplexer 1201 divided by 1024, 4096, 16,384, or 65,536. Thuε, the εignalε on output tapε OSCl and OSC2 are programmed to oεcillate at many different choiceε of frequency. Fig. 11C shows lower right corner tile 115. Corner tile 115 programmably connects long horizontal lines LH0-LH7 and long vertical lines LV0-LV7, respectively, and connects north lines N0-N7 to west lines W0-W7. Corner tile 115 further includes a programmable interconnect structure 1103 which programmably connects a start-up block STARTUP to north lines N0-N7 (and thuε west lines W0-W7) and long vertical lines LV0- LV7 (and thus long horizontal lines LH0-LH7) . Start-up block STARTUP includes circuitry to sequence the signals and control timing of the start-up function as chip 100 (Fig. 1) is activated. During the start-up function, three events are necessary to move from configuration mode to operating mode: release of the signal on a global tri-state signal terminal GTS, release of the signal on a global reset signal terminal GSR, and release of a signal on a load complete terminal DONE (indicating that all configuration bits have been loaded into their appropriate locations in the FPGA) . The start-up block STARTUP allows the user to program the- order in which these signals are released, aε well aε the timing of these signals (for example separating each signal from another εignal by one, two, or three clock cycleε) . Fig. 11D εhows lower left tile 116 with single length and long lines connected similarly to the other three corner tiles. In addition, lower left corner tile 116 includes a read-back unit RDBK. Read-back unit RDBK allowε the user to read the content of the configuration memory onto any data line and out onto any external pin through the data line terminal DATA of readback unit RDBK. The trigger terminal TRIG in read-back unit RDBK carries a signal that triggerε copying of one row of configuration data from the configuration memory into the εame shift register which loaded the configuration memory. The signal on a clock terminal CLK controls shifting out of that data onto line DATA. The εignal on a read-in-progresε terminal RIP preventε the chip from εending another εignal from trigger terminal TRIG while data are εtill being εhifted out. With thiε circuit, depending on the original configuration pathε to corner tile 116, the configuration data for the entire chip iε εhifted out of the chip onto almoεt any one of the external pinε while the chip iε operating. In light of the above description, many other embodimentε of the preεent invention will be apparent to thoεe εkilled in the art. For example, although the above deεcription relates to an embodiment in which core tiles are rectangular or square, another embodiment of the present invention includes tiles having six sides. As mentioned above, core tiles need not be identical. A set of tile designs may be provided which have different logic content from each other. If all tile deεignε follow common boundary conεtraintε, chipε can be formed by combining the tile designs in a variety of patterns. To be succeεεful, each tile design must have a good distribution of signals within the tile. The routing matrix of the tile must efficiently distribute the incoming signals to th logic block input terminalε and take the logic block output εignals to the tile edges. Indeed a chip may be composed in which some tiles include RAM memory and no logic, or a combination of tiles having logic, tiles having memory only, and tiles having routing with no logic or memory. Further, a tile may be designed which includes an input/output pad physically within its structure, and tile designs including a pad may be combined with other tile designε to achieve diεtributed acceεε to logic. Such other embodimentε are intended to fall within the εcope of the preεent invention. The preεent invention iε εet forth in the claimε.
Priority Applications (5)
|Application Number||Priority Date||Filing Date||Title|
|PCT/US1995/001554 WO1995022205A1 (en)||1994-02-15||1995-02-07||Tile based architecture for fpga|
|Publication Number||Publication Date|
|EP0698312A1 true EP0698312A1 (en)||1996-02-28|
Family Applications (1)
|Application Number||Title||Priority Date||Filing Date|
|EP19950909504 Withdrawn EP0698312A1 (en)||1994-02-15||1995-02-07||Tile based architecture for fpga|
Country Status (3)
|EP (1)||EP0698312A1 (en)|
|JP (1)||JP3547446B2 (en)|
|WO (1)||WO1995022205A1 (en)|
Families Citing this family (37)
|Publication number||Priority date||Publication date||Assignee||Title|
|US5543732A (en) *||1995-05-17||1996-08-06||Altera Corporation||Programmable logic array devices with interconnect lines of various lengths|
|US5909126A (en) *||1995-05-17||1999-06-01||Altera Corporation||Programmable logic array integrated circuit devices with interleaved logic array blocks|
|US5900743A (en)||1995-05-17||1999-05-04||Altera Corporation||Programmable logic array devices with interconnect lines of various lengths|
|US5963049A (en) *||1995-05-17||1999-10-05||Altera Corporation||Programmable logic array integrated circuit architectures|
|US5689195A (en)||1995-05-17||1997-11-18||Altera Corporation||Programmable logic array integrated circuit devices|
|US5646545A (en) *||1995-08-18||1997-07-08||Xilinx, Inc.||Time multiplexed programmable logic device|
|US5784313A (en)||1995-08-18||1998-07-21||Xilinx, Inc.||Programmable logic device including configuration data or user data memory slices|
|US5835998A (en) *||1996-04-04||1998-11-10||Altera Corporation||Logic cell for programmable logic devices|
|US5742181A (en) *||1996-06-04||1998-04-21||Hewlett-Packard Co.||FPGA with hierarchical interconnect structure and hyperlinks|
|US5880597A (en) *||1996-09-18||1999-03-09||Altera Corporation||Interleaved interconnect for programmable logic array devices|
|US5999016A (en) *||1996-10-10||1999-12-07||Altera Corporation||Architectures for programmable logic devices|
|US5977793A (en) *||1996-10-10||1999-11-02||Altera Corporation||Programmable logic device with hierarchical interconnection resources|
|US6300794B1 (en)||1996-10-10||2001-10-09||Altera Corporation||Programmable logic device with hierarchical interconnection resources|
|EP0858167A1 (en)||1997-01-29||1998-08-12||Hewlett-Packard Company||Field programmable processor device|
|EP0858168A1 (en) *||1997-01-29||1998-08-12||Hewlett-Packard Company||Field programmable processor array|
|US5982195A (en) *||1997-02-20||1999-11-09||Altera Corporation||Programmable logic device architectures|
|US7148722B1 (en)||1997-02-20||2006-12-12||Altera Corporation||PCI-compatible programmable logic devices|
|US5999015A (en) *||1997-02-20||1999-12-07||Altera Corporation||Logic region resources for programmable logic devices|
|US6127844A (en) *||1997-02-20||2000-10-03||Altera Corporation||PCI-compatible programmable logic devices|
|US6184710B1 (en)||1997-03-20||2001-02-06||Altera Corporation||Programmable logic array devices with enhanced interconnectivity between adjacent logic regions|
|US6121790A (en) *||1997-10-16||2000-09-19||Altera Corporation||Programmable logic device with enhanced multiplexing capabilities in interconnect resources|
|US6107825A (en) *||1997-10-16||2000-08-22||Altera Corporation||Input/output circuitry for programmable logic devices|
|US6107824A (en)||1997-10-16||2000-08-22||Altera Corporation||Circuitry and methods for internal interconnection of programmable logic devices|
|DE69827589T2 (en)||1997-12-17||2005-11-03||Elixent Ltd.||Configurable processing arrangement and method for use of this arrangement is to establish a central unit|
|DE69841256D1 (en)||1997-12-17||2009-12-10||Panasonic Corp||Masking command to command streams forwarded to a processor|
|US6567834B1 (en)||1997-12-17||2003-05-20||Elixent Limited||Implementation of multipliers in programmable arrays|
|US6084427A (en) *||1998-05-19||2000-07-04||Altera Corporation||Programmable logic devices with enhanced multiplexing capabilities|
|JP3576837B2 (en) *||1998-10-30||2004-10-13||日本電気株式会社||Basic cells and the basic cell 2-dimensional array of programmable logic lsi|
|US6215326B1 (en)||1998-11-18||2001-04-10||Altera Corporation||Programmable logic device architecture with super-regions having logic regions and a memory region|
|US6507216B1 (en)||1998-11-18||2003-01-14||Altera Corporation||Efficient arrangement of interconnection resources on programmable logic devices|
|US6407576B1 (en)||1999-03-04||2002-06-18||Altera Corporation||Interconnection and input/output resources for programmable logic integrated circuit devices|
|US6874136B2 (en) *||2002-01-10||2005-03-29||M2000||Crossbar device with reduced parasitic capacitive loading and usage of crossbar devices in reconfigurable circuits|
|US7768314B2 (en)||2004-05-12||2010-08-03||National University Corporation Okayama University||Integrated circuit with multidimensional switch topology|
|US7394287B1 (en) *||2007-05-21||2008-07-01||Altera Corporation||Programmable logic device having complex logic blocks with improved logic cell functionality|
|FR2933826B1 (en) *||2008-07-09||2011-11-18||Univ Paris Curie||programmable logic network, interconnect switch and logic unit for such a network|
|EP2954618A4 (en) *||2013-02-08||2016-10-05||Univ Princeton||Fine-grain dynamically reconfigurable fpga architecture|
|US9239360B2 (en) *||2014-01-28||2016-01-19||Texas Instruments Incorporated||DFT approach to enable faster scan chain diagnosis|
Family Cites Families (5)
|Publication number||Priority date||Publication date||Assignee||Title|
|US5255203A (en) *||1989-08-15||1993-10-19||Advanced Micro Devices, Inc.||Interconnect structure for programmable logic device|
|US5220213A (en) *||1991-03-06||1993-06-15||Quicklogic Corporation||Programmable application specific integrated circuit and logic cell therefor|
|US5241224A (en) *||1991-04-25||1993-08-31||Altera Corporation||High-density erasable programmable logic device architecture using multiplexer interconnections|
|US5260611A (en) *||1991-09-03||1993-11-09||Altera Corporation||Programmable logic array having local and long distance conductors|
|US5260610A (en) *||1991-09-03||1993-11-09||Altera Corporation||Programmable logic element interconnections for programmable logic array integrated circuits|
Non-Patent Citations (1)
|See references of WO9522205A1 *|
Also Published As
|Publication number||Publication date|
|US5815726A (en)||Coarse-grained look-up table architecture|
|US7075333B1 (en)||Programmable circuit optionally configurable as a lookup table or a wide multiplexer|
|US5489857A (en)||Flexible synchronous/asynchronous cell structure for a high density programmable logic device|
|US6580289B2 (en)||Cell architecture to reduce customization in a semiconductor device|
|US5255203A (en)||Interconnect structure for programmable logic device|
|US4918440A (en)||Programmable logic cell and array|
|EP0630115B1 (en)||Configurable logic array|
|US6191610B1 (en)||Method for implementing large multiplexers with FPGA lookup tables|
|US5015884A (en)||Multiple array high performance programmable logic device family|
|EP0213971B1 (en)||Programmable logic array with added array of gates and added output routing flexibility|
|US5764080A (en)||Input/output interface circuitry for programmable logic array integrated circuit devices|
|US5512765A (en)||Extendable circuit architecture|
|DE69721344T2 (en)||FPGA with configurable clock lines|
|EP0591342B1 (en)||Basic cell architecture for mask programmable gate array|
|US7154299B2 (en)||Architecture for programmable logic device|
|US6184713B1 (en)||Scalable architecture for high density CPLDS having two-level hierarchy of routing resources|
|US6137307A (en)||Structure and method for loading wide frames of data from a narrow input bus|
|US7218133B2 (en)||Versatile logic element and logic array block|
|US5635851A (en)||Read and writable data bus particularly for programmable logic devices|
|US6815981B2 (en)||Programmable logic array integrated circuit devices|
|US5220214A (en)||Registered logic macrocell with product term allocation and adjacent product term stealing|
|US5089973A (en)||Programmable logic cell and array|
|US5485104A (en)||Logic allocator for a programmable logic device|
|EP0569137B1 (en)||Programmable logic array integrated circuit|
|US5936424A (en)||High speed bus with tree structure for selecting bus driver|
|AK||Designated contracting states:||
Kind code of ref document: A1
Designated state(s): DE FR GB
|17P||Request for examination filed||
Effective date: 19951204
|18D||Deemed to be withdrawn||
Effective date: 19970902