Patent Grant
8269523
Multi-stage interconnection networks such as Benes networks and butterfly fat tree networks are widely useful in telecommunications, parallel and distributed computing. However VLSI layouts, known in the prior art, of these interconnection networks in an integrated circuit are inefficient and complicated.
Other multi-stage interconnection networks including butterfly fat tree networks, Banyan networks, Batcher-Banyan networks, Baseline networks, Delta networks, Omega networks and Flip networks have been widely studied particularly for self routing packet switching applications. Also Benes Networks with radix of two have been widely studied and it is known that Benes Networks of radix two are shown to be built with back to back baseline networks which are rearrangeably nonblocking for unicast connections.
The most commonly used VLSI layout in an integrated circuit is based on a two-dimensional grid model comprising only horizontal and vertical tracks. An intuitive interconnection network that utilizes two-dimensional grid model is 2D Mesh Network and its variations such as segmented mesh networks. Hence routing networks used in VLSI layouts are typically 2D mesh networks and its variations. However Mesh Networks require large scale cross points typically with a growth rate of O(N2) where N is the number of computing elements, ports, or logic elements depending on the application.
Multi-stage interconnection with a growth rate of O(N×log N) requires significantly small number of cross points. U.S. Pat. No. 6,185,220 entitled “Grid Layouts of Switching and Sorting Networks” granted to Muthukrishnan et al. describes a VLSI layout using existing VLSI grid model for Benes and Butterfly networks. U.S. Pat. No. 6,940,308 entitled “Interconnection Network for a Field Programmable Gate Array” granted to Wong describes a VLSI layout where switches belonging to lower stage of Benes Network are layed out close to the logic cells and switches belonging to higher stages are layed out towards the center of the layout.
Due to the inefficient and in some cases impractical VLSI layout of Benes and butterfly fat tree networks on a semiconductor chip, today mesh networks and segmented mesh networks are widely used in the practical applications such as field programmable gate arrays (FPGAs), programmable logic devices (PLDs), and parallel computing interconnects. The prior art VLSI layouts of Benes and butterfly fat tree networks and VLSI layouts of mesh networks and segmented mesh networks require large area to implement the switches on the chip, large number of wires, longer wires, with increased power consumption, increased latency of the signals which effect the maximum clock speed of operation. Some networks may not even be implemented practically on a chip due to the lack of efficient layouts.
When large scale sub-integrated circuit blocks with inlet and outlet links are layed out in an integrated circuit device in a two-dimensional grid arrangement, (for example in an FPGA where the sub-integrated circuit blocks are Lookup Tables) the most intuitive routing network is a network that uses horizontal and vertical links only (the most often used such a network is one of the variations of a 2D Mesh network). A direct embedding of a generalized multi-stage network on to a 2D Mesh network is neither simple nor efficient.
In accordance with the invention, VLSI layouts of generalized multi-stage networks for broadcast, unicast and multicast connections are presented using only horizontal and vertical links The VLSI layouts employ shuffle exchange links where outlet links of cross links from switches in a stage in one sub-integrated circuit block are connected to inlet links of switches in the succeeding stage in another sub-integrated circuit block so that said cross links are either vertical links or horizontal and vice versa. In one embodiment the sub-integrated circuit blocks are arranged in a hypercube arrangement in a two-dimensional plane. The VLSI layouts exploit the benefits of significantly lower cross points, lower signal latency, lower power and full connectivity with significantly fast compilation.
The VLSI layouts presented are applicable to generalized multi-stage networks V(N1, N2, d, s), generalized folded multi-stage networks Vfold(N1, N2, d, s), generalized butterfly fat tree networks Vbft(N1, N2, d, s), generalized multi-link multi-stage networks Vmlink(N1, N2, d, s), generalized folded multi-link multi-stage networks Vfold-mlink(N1, N2, d, s), generalized multi-link butterfly fat tree networks Vmlink-bft(N1, N2, d, s), and generalized hypercube networks Vhcube(N1, N2, d, s) for s=1, 2, 3 or any number in general. The embodiments of VLSI layouts are useful in wide target applications such as FPGAs, CPLDs, pSoCs, ASIC placement and route tools, networking applications, parallel & distributed computing, and reconfigurable computing.
FIG. 1K1 is a diagram 100M1 detailed connections of BLOCK 1_2 in the network layout 100C in one embodiment, illustrating the connection links going in and coming out when the layout 100C is implementing V(N, d, s) or Vfold(N, d, s) for s=1.
FIG. 1L1 is a diagram 100L1 detailed connections of BLOCK 1_2 in the network layout 100C in one embodiment, illustrating the connection links going in and coming out when the layout 100C is implementing V(N, d, s) or Vfold(N, d, s) for s=1.
FIG. 2A1 is a diagram 200A1 of an exemplary symmetrical multi-link multi-stage network Vfold-mlink(N, d, s) having inverse Benes connection topology of one stage with N=2, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention. FIG. 2A2 is a diagram 200A2 of the equivalent symmetrical folded multi-link multi-stage network Vfold-mlink(N, d, s) of the network 200A1 shown in FIG. 2A1, having inverse Benes connection topology of one stage with N=2, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention. FIG. 2A3 is a diagram 200A3 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2A2, in one embodiment, illustrating all the connection links.
FIG. 2B1 is a diagram 200B1 of an exemplary symmetrical multi-link multi-stage network Vfold-mlink(N, d, s) having inverse Benes connection topology of one stage with N=4, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention. FIG. 2B2 is a diagram 200B2 of the equivalent symmetrical folded multi-link multi-stage network Vfold-mlink(N, d, s) of the network 200B1 shown in FIG. 2B1, having inverse Benes connection topology of one stage with N=4, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention. FIG. 2B3 is a diagram 200B3 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2B2, in one embodiment, illustrating the connection links belonging with in each block only. FIG. 2B4 is a diagram 200B4 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2B2, in one embodiment, illustrating the connection links ML(1,i) for i=[1, 8] and ML(2,i) for i=[1,8].
FIG. 2C11 is a diagram 200C11 of an exemplary symmetrical multi-link multi-stage network Vfold-mlink(N, d, s) having inverse Benes connection topology of one stage with N=8, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention. FIG. 2C12 is a diagram 200C12 of the equivalent symmetrical folded multi-link multi-stage network Vfold-mlink(N, d, s) of the network 200C11 shown in FIG. 2C11, having inverse Benes connection topology of one stage with N=8, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention.
FIG. 2C21 is a diagram 200C21 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2C12, in one embodiment, illustrating the connection links belonging with in each block only. FIG. 2C22 is a diagram 200C22 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2C12, in one embodiment, illustrating the connection links ML(1,i) for i=[1, 16] and ML(4,i) for i=[1,16]. FIG. 2C23 is a diagram 200C23 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2C12, in one embodiment, illustrating the connection links ML(2,i) for i=[1, 16] and ML(3,i) for i=[1,16].
FIG. 2D1 is a diagram 200D1 of an exemplary symmetrical multi-link multi-stage network Vfold-mlink(N, d, s) having inverse Benes connection topology of one stage with N=16, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention.
FIG. 2D2 is a diagram 200D2 of the equivalent symmetrical folded multi-link multi-stage network Vfold-mlink(N, d, s) of the network 200D1 shown in FIG. 2D1, having inverse Benes connection topology of one stage with N=16, d=2 and s=2, strictly nonblocking network for unicast connections and rearrangeably nonblocking network for arbitrary fan-out multicast connections, in accordance with the invention.
FIG. 2D3 is a diagram 200D3 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2D2, in one embodiment, illustrating the connection links belonging with in each block only.
FIG. 2D4 is a diagram 200D4 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2D2, in one embodiment, illustrating the connection links ML(1,i) for i=[1, 32] and ML(6,i) for i=[1,32].
FIG. 2D5 is a diagram 200D5 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2D2, in one embodiment, illustrating the connection links ML(2,i) for i=[1, 32] and ML(5,i) for i=[1,32].
FIG. 2D6 is a diagram 200D6 layout of the network Vfold-mlink(N, d, s) shown in FIG. 2D2, in one embodiment, illustrating the connection links ML(3,i) for i=[1, 32] and ML(4,i) for i=[1,32].
FIG. 4C1 is a diagram 400C1 layout of the network Vfold-mlink(N, d, s) shown in
FIG. 5A1 is a diagram 500A1 of an exemplary prior art implementation of a two by two switch; FIG. 5A2 is a diagram 500A2 for programmable integrated circuit prior art implementation of the diagram 500A1 of FIG. 5A1; FIG. 5A3 is a diagram 500A3 for one-time programmable integrated circuit prior art implementation of the diagram 500A1 of FIG. 5A1; FIG. 5A4 is a diagram 500A4 for integrated circuit placement and route implementation of the diagram 500A1 of FIG. 5A1.
The present invention is concerned with the VLSI layouts of arbitrarily large switching networks for broadcast, unicast and multicast connections. Particularly switching networks considered in the current invention include: generalized multi-stage networks V(N1, N2, d, s), generalized folded multi-stage networks Vfold(N1, N2, d, s), generalized butterfly fat tree networks Vbft(N1, N2, d, s), generalized multi-link multi-stage networks Vmlink(N1, N2, d, s), generalized folded multi-link multi-stage networks Vfold-mlink(N1, N2, d, s), generalized multi-link butterfly fat tree networks Vmlink-bft(N1, N2, d, s), and generalized hypercube networks Vhcube(N1, N2, d, s) for s=1, 2, 3 or any number in general.
Efficient VLSI layout of networks on a semiconductor chip are very important and greatly influence many important design parameters such as the area taken up by the network on the chip, total number of wires, length of the wires, latency of the signals, capacitance and hence the maximum clock speed of operation. Some networks may not even be implemented practically on a chip due to the lack of efficient layouts. The different varieties of multi-stage networks described above have not been implemented previously on the semiconductor chips efficiently. For example in Field Programmable Gate Array (FPGA) designs, multi-stage networks described in the current invention have not been successfully implemented primarily due to the lack of efficient VLSI layouts. Current commercial FPGA products such as Xilinx Vertex, Altera's Stratix implement island-style architecture using mesh and segmented mesh routing interconnects using either full crossbars or sparse crossbars. These routing interconnects consume large silicon area for crosspoints, long wires, large signal propagation delay and hence consume lot of power.
The current invention discloses the VLSI layouts of numerous types of multi-stage networks which are very efficient. Moreover they can be embedded on to mesh and segmented mesh routing interconnects of current commercial FPGA products. The VLSI layouts disclosed in the current invention are applicable to including the numerous generalized multi-stage networks disclosed in the following patent applications, filed concurrently:
1) Strictly and rearrangeably nonblocking for arbitrary fan-out multicast and unicast for generalized multi-stage networks V(N1, N2, d, s) with numerous connection topologies and the scheduling methods are described in detail in the PCT Application Serial No. PCT/US08/56064 that is incorporated by reference above.
2) Strictly and rearrangeably nonblocking for arbitrary fan-out multicast and unicast for generalized butterfly fat tree networks Vbft(N1, N2, d, s) with numerous connection topologies and the scheduling methods are described in detail in PCT Application Serial No. PCT/US08/64603 that is incorporated by reference above.
3) Rearrangeably nonblocking for arbitrary fan-out multicast and unicast, and strictly nonblocking for unicast for generalized multi-link multi-stage networks Vmlink(N1, N2, d, s) and generalized folded multi-link multi-stage networks Vfold-mlink(N1, N2, d, s) with numerous connection topologies and the scheduling methods are described in detail in PCT Application Serial No. PCT/US08/64604 that is incorporated by reference above.
4) Strictly and rearrangeably nonblocking for arbitrary fan-out multicast and unicast for generalized multi-link butterfly fat tree networks Vmlink-bft(N1, N2, d, s) with numerous connection topologies and the scheduling methods are described in detail in PCT Application Serial No. PCT/US08/64603 that is incorporated by reference above.
5) Strictly and rearrangeably nonblocking for arbitrary fan-out multicast and unicast for generalized folded multi-stage networks Vfold(N1, N2, d, s) with numerous connection topologies and the scheduling methods are described in detail in PCT Application Serial No. PCT/US08/64604 that is incorporated by reference above.
6) Strictly nonblocking for arbitrary fan-out multicast for generalized multi-link multi-stage networks Vmlink(N1, N2, d, s) and generalized folded multi-link multi-stage networks Vfold-mlink(N1, N2, d, s) with numerous connection topologies and the scheduling methods are described in detail in PCT Application Serial No. PCT/US08/64604 that is incorporated by reference above.
7) VLSI layouts of numerous types of multi-stage networks with locality exploitation are described in U.S. Provisional Patent Application Ser. No. 61/252,603 that is incorporated by reference above.
8) VLSI layouts of numerous types of multistage pyramid networks are described in U.S. Provisional Patent Application Ser. No. 61/252,609 that is incorporated by reference above.
In addition the layouts of the current invention are also applicable to generalized multi-stage pyramid networks Vp(N1, N2, d, s), generalized folded multi-stage pyramid networks Vfold-p(N1, N2, d, s), generalized butterfly fat pyramid networks Vbfp(N1, N2 d, s), generalized multi-link multi-stage pyramid networks Vmlink-p(N1, N2, d, s), generalized folded multi-link multi-stage pyramid networks Vfold-mlink-p(N1, N2, d, s), generalized multi-link butterfly fat pyramid networks Vmlink-bfp(N1, N2, d, s), and generalized hypercube networks Vhcube(N1, N2, d, s) for s=1, 2, 3 or any number in general.
Symmetric RNB Generalized Multi-Link Multi-Stage Network Vmlink(N1, N2, d, s):
Referring to diagram 100A in
As disclosed in PCT Application Serial No. PCT/US08/64604 that is incorporated by reference above, such a network can be operated in rearrangeably non-blocking manner for arbitrary fan-out multicast connections and also can be operated in strictly non-blocking manner for unicast connections.
In one embodiment of this network each of the input switches IS1-IS4 and output switches OS1-OS4 are crossbar switches. The number of switches of input stage 110 and of output stage 120 can be denoted in general with the variable
where N is the total number of inlet links or outlet links. The number of middle switches in each middle stage is denoted by
The size of each input switch IS1-IS4 can be denoted in general with the notation d*2d and each output switch OS1-OS4 can be denoted in general with the notation 2d*d. Likewise, the size of each switch in any of the middle stages can be denoted as 2d*2d. A switch as used herein can be either a crossbar switch, or a network of switches each of which in turn may be a crossbar switch or a network of switches. A symmetric multi-stage network can be represented with the notation Vmlink(N, d, s), where N represents the total number of inlet links of all input switches (for example the links IL1-IL32), d represents the inlet links of each input switch or outlet links of each output switch, and s is the ratio of number of outgoing links from each input switch to the inlet links of each input switch.
Each of the
input switches IS1-IS16 are connected to exactly d switches in middle stage 130 through two links each for a total of 2×d links (for example input switch IS1 is connected to middle switch MS(1,1) through the links ML(1,1), ML(1,2), and also connected to middle switch MS(1,2) through the links ML(1,3) and ML(1,4)). The middle links which connect switches in the same row in two successive middle stages are called hereinafter straight middle links; and the middle links which connect switches in different rows in two successive middle stages are called hereinafter cross middle links. For example, the middle links ML(1,1) and ML(1,2) connect input switch IS1 and middle switch MS(1,1), so middle links ML(1,1) and ML(1,2) are straight middle links; where as the middle links ML(1,3) and ML(1,4) connect input switch IS1 and middle switch MS(1,2), since input switch IS1 and middle switch MS(1,2) belong to two different rows in diagram 100A of
Each of the
middle switches MS(1,1)-MS(1,16) in the middle stage 130 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(1,1) and ML(1,2) are connected to the middle switch MS(1,1) from input switch IS1, and the links ML(1,7) and ML(1,8) are connected to the middle switch MS(1,1) from input switch IS2) and also are connected to exactly d switches in middle stage 140 through two links each for a total of 2×d links (for example the links ML(2,1) and ML(2,2) are connected from middle switch MS(1,1) to middle switch MS(2,1), and the links ML(2,3) and ML(2,4) are connected from middle switch MS(1,1) to middle switch MS(2,3)).
Each of the
middle switches MS(2,1)-MS(2,16) in the middle stage 140 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(2,1) and ML(2,2) are connected to the middle switch MS(2,1) from input switch MS(1,1), and the links ML(1,11) and ML(1,12) are connected to the middle switch MS(2,1) from input switch MS(1,3)) and also are connected to exactly d switches in middle stage 150 through two links each for a total of 2×d links (for example the links ML(3,1) and ML(3,2) are connected from middle switch MS(2,1) to middle switch MS(3,1), and the links ML(3,3) and ML(3,4) are connected from middle switch MS(2,1) to middle switch MS(3,5)).
Each of the
middle switches MS(3,1)-MS(3,16) in the middle stage 150 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(3,1) and ML(3,2) are connected to the middle switch MS(3,1) from input switch MS(2,1), and the links ML(2,19) and ML(2,20) are connected to the middle switch MS(3,1) from input switch MS(2,5)) and also are connected to exactly d switches in middle stage 160 through two links each for a total of 2×d links (for example the links ML(4,1) and ML(4,2) are connected from middle switch MS(3,1) to middle switch MS(4,1), and the links ML(4,3) and ML(4,4) are connected from middle switch MS(3,1) to middle switch MS(4,9)).
Each of the
middle switches MS(4,1)-MS(4,16) in the middle stage 160 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(4,1) and ML(4,2) are connected to the middle switch MS(4,1) from input switch MS(3,1), and the links ML(4,35) and ML(4,36) are connected to the middle switch MS(4,1) from input switch MS(3,9)) and also are connected to exactly d switches in middle stage 170 through two links each for a total of 2×d links (for example the links ML(5,1) and ML(5,2) are connected from middle switch MS(4,1) to middle switch MS(5,1), and the links ML(5,3) and ML(5,4) are connected from middle switch MS(4,1) to middle switch MS(5,9)).
Each of the
middle switches MS(5,1)-MS(5,16) in the middle stage 170 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(5,1) and ML(5,2) are connected to the middle switch MS(5,1) from input switch MS(4,1), and the links ML(5,35) and ML(5,36) are connected to the middle switch MS(5,1) from input switch MS(4,9)) and also are connected to exactly d switches in middle stage 180 through two links each for a total of 2×d links (for example the links ML(6,1) and ML(6,2) are connected from middle switch MS(5,1) to middle switch MS(6,1), and the links ML(6,3) and ML(6,4) are connected from middle switch MS(5,1) to middle switch MS(6,5)).
Each of the
middle switches MS(6,1)-MS(6,16) in the middle stage 180 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(6,1) and ML(6,2) are connected to the middle switch MS(6,1) from input switch MS(5,1), and the links ML(6,19) and ML(6,20) are connected to the middle switch MS(6,1) from input switch MS(5,5)) and also are connected to exactly d switches in middle stage 190 through two links each for a total of 2×d links (for example the links ML(7,1) and ML(7,2) are connected from middle switch MS(6,1) to middle switch MS(7,1), and the links ML(7,3) and ML(7,4) are connected from middle switch MS(6,1) to middle switch MS(7,3)).
Each of the
middle switches MS(7,1)-MS(7,16) in the middle stage 190 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(7,1) and ML(7,2) are connected to the middle switch MS(7,1) from input switch MS(6,1), and the links ML(7,11) and ML(7,12) are connected to the middle switch MS(7,1) from input switch MS(6,3)) and also are connected to exactly d switches in middle stage 120 through two links each for a total of 2×d links (for example the links ML(8,1) and ML(8,2) are connected from middle switch MS(7,1) to middle switch MS(8,1), and the links ML(8,3) and ML(8,4) are connected from middle switch MS(7,1) to middle switch OS2).
Each of the
middle switches OS1-OS16 in the middle stage 120 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(8,1) and ML(8,2) are connected to the output switch OS1 from input switch MS(7,1), and the links ML(8,7) and ML(7,8) are connected to the output switch OS1 from input switch MS(7,2)).
Finally the connection topology of the network 100A shown in
Referring to diagram 100B in
Middle stage 130 and middle stage 190 are placed together. That is middle switches MS(1,1) and MS(7,1) are placed together, middle switches MS(1,2) and MS(7,2) are placed together, and similarly middle switches MS(1,16) and MS(7,16) are placed together. All the right going middle links {i.e., middle links ML(1,1)-ML(1,64) and middle links ML(2,1)-ML(2,64)} correspond to middle switches MS(1,1)-MS(1,16), and all the left going middle links {i.e., middle links ML(7,1)-ML(7,64) and middle links ML(8,1) and ML(8,64)} correspond to middle switches MS(7,1)-MS(7,16).
Middle stage 140 and middle stage 180 are placed together. That is middle switches MS(2,1) and MS(6,1) are placed together, middle switches MS(2,2) and MS(6,2) are placed together, and similarly middle switches MS(2,16) and MS(6,16) are placed together. All the right going middle links {i.e., middle links ML(2,1)-ML(2,64) and middle links ML(3,1)-ML(3,64)} correspond to middle switches MS(2,1)-MS(2,16), and all the left going middle links {i.e., middle links ML(6,1)-ML(6,64) and middle links ML(7,1) and ML(7,64)} correspond to middle switches MS(6,1)-MS(6,16).
Middle stage 150 and middle stage 170 are placed together. That is middle switches MS(3,1) and MS(5,1) are placed together, middle switches MS(3,2) and MS(5,2) are placed together, and similarly middle switches MS(3,16) and MS(5,16) are placed together. All the right going middle links {i.e., middle links ML(3,1)-ML(3,64) and middle links ML(4,1)-ML(4,64)} correspond to middle switches MS(3,1)-MS(3,16), and all the left going middle links {i.e., middle links ML(5,1)-ML(5,64) and middle links ML(6,1) and ML(6,64)} correspond to middle switches MS(5,1)-MS(5,16).
Middle stage 160 is placed alone. All the right going middle links are the middle links ML(4,1)-ML(4,64) and all the left going middle links are middle links ML(5,1)-ML(5,64).
In one embodiment, in the network 100B of
Hypercube Topology Layout Schemes:
Referring to layout 100C of
All the straight middle links are illustrated in layout 100C of
Even though it is not illustrated in layout 100C of
Recursively in each quadrant there are four sub-quadrants. For example in top-left quadrant there are four sub-quadrants namely top-left sub-quadrant, bottom-left sub-quadrant, top-right sub-quadrant and bottom-right sub-quadrant. Top-left sub-quadrant of top-left quadrant implements Block 1_2. Bottom-left sub-quadrant of top-left quadrant implements Block 3_4. Top-right sub-quadrant of top-left quadrant implements Block 5_6. Finally bottom-right sub-quadrant of top-left quadrant implements Block 7_8. Similarly there are two sub-halves in each quadrant. For example in top-left quadrant there are two sub-halves namely left-sub-half and right-sub-half. Left-sub-half of top-left quadrant implements Block 1_2 and Block 3_4. Right-sub-half of top-left quadrant implements Block 5_6 and Block 7_8. Finally applicant notes that in each quadrant or half the blocks are arranged as a general binary hypercube. Recursively in larger multi-stage network Vfold-mlink(N1, N2, d, s) where N1=N2>32, the layout in this embodiment in accordance with the current invention, will be such that the super-quadrants will also be arranged in d-ary hypercube manner. (In the embodiment of the layout 100C of
Layout 100D of
Layout 100E of
Layout 100F of
Layout 100G of
The complete layout for the network 100B of
Some of the key aspects of the current invention are discussed. 1) All the switches in one row of the multi-stage network 100B are implemented in a single block. 2) The blocks are placed in such a way that all the inter-block links are either horizontal tracks or vertical tracks; 3) Since all the inter-block links are either horizontal or vertical tracks, all the inter-block links can be mapped on to island-style architectures in current commercial FPGA's; 4) The length of the longest wire is about half of the width (or length) of the complete layout (For example middle link ML(4,4) is about half the width of the complete layout).
In accordance with the current invention, the layout 100C in
Bottom-left super-quadrant implements the blocks from block 33_34 to block 63_64. Top-right super-quadrant implements the blocks from block 65_66 to block 95_96. And bottom-right super-quadrant implements the blocks from block 97_98 to block 127_128. In all these three super-quadrants also, the inter-block link connection topology is the exactly the same between the switches 1 and 2; switches 2 and 3; switches 3 and 4; switches 4 and 5 as that of the top-left super-quadrant.
Recursively in accordance with the current invention, the inter-block links connecting the switch 5 and switch 6 will be vertical tracks between the corresponding switches of top-left super-quadrant and bottom-left super-quadrant. And similarly the inter-block links connecting the switch 5 and switch 6 will be vertical tracks between the corresponding switches of top-right super-quadrant and bottom-right super-quadrant. The inter-block links connecting the switch 6 and switch 7 will be horizontal tracks between the corresponding switches of top-left super-quadrant and top-right super-quadrant. And similarly the inter-block links connecting the switch 6 and switch 7 will be horizontal tracks between the corresponding switches of bottom-left super-quadrant and bottom-right super-quadrant.
Referring to diagram 100I of
That is the switches that are placed together in Block 1_2 as shown in
Input switch IS1 is implemented as two by four switch with the inlet links IL1 and IL2 being the inputs of the input switch IS1 and middle links ML(1,1)-ML(1,4) being the outputs of the input switch IS1; and output switch OS1 is implemented as four by two switch with the middle links ML(8,1)-ML(8,4) being the inputs of the output switch OS1 and outlet links OL1-OL2 being the outputs of the output switch OS1.
Middle switch MS(1,1) is implemented as four by four switch with the middle links ML(1,1), ML(1,2), ML(1,7) and ML(1,8) being the inputs and middle links ML(2,1)-ML(2,4) being the outputs; and middle switch MS(7,1) is implemented as four by four switch with the middle links ML(7,1), ML(7,2), ML(7,11) and ML(7,12) being the inputs and middle links ML(8,1)-ML(8,4) being the outputs. Similarly all the other middle switches are also implemented as four by four switches as illustrated in 100I of
Now the VLSI layouts of generalized multi-link multi-stage network Vmlink(N1, N2, d, s) where N1=N2<32; d=2; s=2 and its corresponding version of folded generalized multi-link multi-stage network Vfold-mlink(N1, N2, d, s) where N1=N2<32; d=2; s=2 are discussed. Referring to diagram 200A1 of FIG. 2A1 is generalized multi-link multi-stage network Vmlink(N1, N2, d, s) where N1=N2=2; d=2. Diagram 200A2 of FIG. 2A2 illustrates the corresponding folded generalized multi-link multi-stage network Vfold-mlink(N1, N2, d, s) where N1=N2=2; d=2, version of the diagram 200A1 of FIG. 2A1. Layout 200A3 of FIG. 2A3 illustrates the VLSI layout of the network 200A2 of FIG. 2A2. There is only one block i.e., Block 1_2 comprising switch 1. Just like in the layout 100C of
Referring to diagram 200B1 of FIG. 2B1 is generalized multi-link multi-stage network Vmlink(N1, N2, d, s) where N1=N2=4; d=2; s=2. Diagram 200B2 of FIG. 2B2 illustrates the corresponding folded generalized multi-link multi-stage network Vfold-mlink(N1, N2, d, s) where N1=N2=4; d=2; s=2, version of the diagram 200B1 of FIG. 2B1. Layout 200B3 of FIG. 2B3 illustrates the VLSI layout of the network 200B2 of FIG. 2B2. There are two blocks i.e., Block 1_2 and Block 3_4 each comprising switch 1 and switch 2. Switch 1 in each block consists of the corresponding input switch and output switch. For example switch 1 in Block 1_2 consists of input switch IS1 and output switch OS1. Similarly switch 2 in Block 1_2 consists of middle switch (1,1). Layout 200B4 of FIG. 2B4 illustrates the inter-block links of the VLSI layout diagram 200B3 of FIG. 2B3. For example middle links ML(1,4) and ML(2,8). It must be noted that all the inter-block links are vertical tracks in this layout. (Alternatively all the inter-blocks can also be implemented as horizontal tracks).
Referring to diagram 200C11 of FIG. 2C11 is generalized multi-link multi-stage network Vmlink(N1, N2, d, s) where N1=N2=8; d=2; s=2. Diagram 200C12 of FIG. 2C12 illustrates the corresponding folded generalized multi-link multi-stage network Vfold-mlink(N1, N2, d, s) where N1=N2=8; d=2; s=2, version of the diagram 200C11 of FIG. 2C11. Layout 200C21 of FIG. 2C21 illustrates the VLSI layout of the network 200C12 of FIG. 2C12. There are four blocks i.e., Block 1_2, Block 3_4, Block 5_6, and Block 7_8 each comprising switch 1, switch 2 and switch 3. For example switch 1 in Block 1_2 consists of input switch IS1 and output switch OS1; Switch 2 in Block 1_2 consists of MS(1,1) and MS(3,1). Switch 3 in Block 1_2 consists of MS(2,1).
Layout 200C22 of FIG. 2C22 illustrates the inter-block links between the switch 1 and switch 2 of the VLSI layout diagram 200C21 of FIG. 2C21. For example middle links ML(1,4) and ML(4,8) are connected between Block 1_2 and Block 3_4. It must be noted that all the inter-block links between switch 1 and switch 2 of all blocks are vertical tracks in this layout. Layout 200C23 of FIG. 2C23 illustrates the inter-block links between the switch 2 and switch 3 of the VLSI layout diagram 200C21 of FIG. 2C21. For example middle links ML(2,12) and ML(3,4) are connected between Block 1_2 and Block 5_6. It must be noted that all the inter-block links between switch 2 and switch 3 of all blocks are horizontal tracks in this layout
Referring to diagram 200D1 of FIG. 2D1 is generalized multi-link multi-stage network Vmlink(N1, N2, d, s) where N1=N2=16; d=2; s=2. Diagram 200D2 of FIG. 2D2 illustrates the corresponding folded generalized multi-link multi-stage network Vfold-mlink(N1, N2, d, s) where N1=N2=16; d=2; s=2, version of the diagram 200D1 of FIG. 2D1. Layout 200D3 of FIG. 2D3 illustrates the VLSI layout of the network 200D2 of FIG. 2D2. There are eight blocks i.e., Block 1_2, Block 3_4, Block 5_6, Block 7_8, Block 9_10, Block 11_12, Block 13_14 and Block 15_16 each comprising switch 1, switch 2, switch 3 and switch 4. For example switch 1 in Block 1_2 consists of input switch IS1 and output switch OS1; Switch 2 in Block 1_2 consists of MS(1,1) and MS(5,1). Switch 3 in Block 1_2 consists of MS(2,1) and MS(4,1), and switch 4 in Block 1_2 consists of MS(3,1).
Layout 200D4 of FIG. 2D4 illustrates the inter-block links between the switch 1 and switch 2 of the VLSI layout diagram 200D3 of FIG. 2D3. For example middle links ML(1,4) and ML(6,8) are connected between Block 1_2 and Block 3_4. It must be noted that all the inter-block links between switch 1 and switch 2 of all blocks are vertical tracks in this layout. Layout 200D5 of FIG. 2D5 illustrates the inter-block links between the switch 2 and switch 3 of the VLSI layout diagram 200D3 of FIG. 2D3. For example middle links ML(2,12) and ML(5,4) are connected between Block 1_2 and Block 5_6. It must be noted that all the inter-block links between switch 2 and switch 3 of all blocks are horizontal tracks in this layout. Layout 200D6 of FIG. 2D6 illustrates the inter-block links between the switch 3 and switch 4 of the VLSI layout diagram 200D3 of FIG. 2D3. For example middle links ML(3,4) and ML(4,20) are connected between Block 1_2 and Block 9_10. It must be noted that all the inter-block links between switch 3 and switch 4 of all blocks are vertical tracks in this layout.
In another embodiment in the network 100B of
Layout diagrams 100C in
Referring to diagram 100J of
That is the switches that are placed together in Block 1_2 as shown in
Combined input and output switch IS1&OS1 is implemented as six by six switch with the inlet links IL1, IL2 and ML(8,1)-ML(8,4) being the inputs and middle links ML(1,1)-ML(1,4), and outlet links OL1-OL2 being the outputs.
Middle switch MS(1,1) is implemented as eight by eight switch with the middle links ML(1,1), ML(1,2), ML(1,7), ML(1,8), ML(7,1), ML(7,2), ML(7,11) and ML(7,12) being the inputs and middle links ML(2,1)-ML(2,4) and middle links ML(8,1)-ML(8,4) being the outputs. Similarly all the other middle switches are also implemented as eight by eight switches as illustrated in 100J of
In another embodiment, middle switch MS(1,1) (or the middle switches in any of the middle stage excepting the root middle stage) of Block 1_2 of Vmlink-bft(N1, N2, d, s) can be implemented as a four by eight switch and a four by four switch to save cross points. This is because the left going middle links of these middle switches are never setup to the right going middle links. For example, in middle switch MS(1,1) of Block 1_2 as shown
In one embodiment, in the network 100B of
The switches, corresponding to the middle stages that are placed together are implemented as four two by two switches. For example middle switches MS(1,1), MS(1,17), MS(7,1), and MS(7,17) are placed together; so middle switch MS(1,1) is implemented as two by two switch with middle links ML(1,1) and ML(1,7) being the inputs and middle links ML(2,1) and ML(2,3) being the outputs; middle switch MS(1,17) is implemented as two by two switch with the middle links ML(1,2) and ML(1,8) being the inputs and middle links ML(2,2) and ML(2,4) being the outputs; middle switch MS(7,1) is implemented as two by two switch with middle links ML(7,1) and ML(7,11) being the inputs and middle links ML(8,1) and ML(8,3) being the outputs; And middle switch MS(7,17) is implemented as two by two switch with the middle links ML(7,2) and ML(7,12) being the inputs and middle links ML(8,2) and ML(8,4) being the outputs; Similarly in this embodiment of network 100B all the switches that are placed together are implemented as separate switches.
Layout diagrams 100C in
Referring to diagram 100K of
That is the switches that are placed together in Block 1_2 as shown in
Input switch IS1 and output switch OS1 are placed together; so input switch IS1 is implemented as two by four switch with the inlet links IL1 and IL2 being the inputs and middle links ML(1,1)-ML(1,4) being the outputs; and output switch OS1 is implemented as four by two switch with the middle links ML(8,1), ML(8,4), ML(8,7) and ML(8,8) being the inputs and outlet links OL1-OL2 being the outputs.
Middle switches MS(1,1), MS(1,17), MS(7,1), and MS(7,17) are placed together; so middle switch MS(1,1) is implemented as two by two switch with middle links ML(1,1) and ML(1,7) being the inputs and middle links ML(2,1) and ML(2,3) being the outputs; middle switch MS(1,17) is implemented as two by two switch with the middle links ML(1,2) and ML(1,8) being the inputs and middle links ML(2,2) and ML(2,4) being the outputs; middle switch MS(7,1) is implemented as two by two switch with middle links ML(7,1) and ML(7,11) being the inputs and middle links ML(8,1) and ML(8,3) being the outputs; And middle switch MS(7,17) is implemented as two by two switch with the middle links ML(7,2) and ML(7,12) being the inputs and middle links ML(8,2) and ML(8,4) being the outputs Similarly all the other middle switches are also implemented as two by two switches as illustrated in 100K of
In one embodiment, in the network 100B of
The switches, corresponding to the middle stages that are placed together are implemented as two, two by two switches. For example middle switches MS(1,1) and MS(7,1) are placed together; so middle switch MS(1,1) is implemented as two by two switch with middle links ML(1,1) and ML(1,3) being the inputs and middle links ML(2,1) and ML(2,2) being the outputs; middle switch MS(7,1) is implemented as two by two switch with middle links ML(7,1) and ML(7,5) being the inputs and middle links ML(8,1) and ML(8,2) being the outputs; Similarly in this embodiment of network 100B all the switches that are placed together are implemented as two separate switches.
Layout diagrams 100C in
Referring to diagram 100K1 of FIG. 1K1 illustrates a high-level implementation of Block 1_2 (Each of the other blocks have similar implementation) for the layout 100C of
That is the switches that are placed together in Block 1_2 as shown in FIG. 1K1 are namely the input switch IS1 and output switch OS1 belonging to switch 1, illustrated by dotted lines, (as noted before switch 1 is for illustration purposes only, in practice the switches implemented are input switch IS1 and output switch OS1); middle switches MS(1,1) and MS(7,1) belonging to switch 2; middle switches MS(2,1) and MS(6,1) belonging to switch 3; middle switches MS(3,1) and MS(5,1) belonging to switch 4; And middle switch MS(4,1) belonging to switch 5.
Input switch IS1 and output switch OS1 are placed together; so input switch IS1 is implemented as two by two switch with the inlet links IL1 and IL2 being the inputs and middle links ML(1,1)-ML(1,2) being the outputs; and output switch OS1 is implemented as two by two switch with the middle links ML(8,1) and ML(8,3) being the inputs and outlet links OL1-OL2 being the outputs.
Middle switches MS(1,1) and MS(7,1) are placed together; so middle switch MS(1,1) is implemented as two by two switch with middle links ML(1,1) and ML(1,3) being the inputs and middle links ML(2,1) and ML(2,2) being the outputs; And middle switch MS(7,1) is implemented as two by two switch with middle links ML(7,1) and ML(7,5) being the inputs and middle links ML(8,1) and ML(8,2) being the outputs. Similarly all the other middle switches are also implemented as two by two switches as illustrated in 100K1 of FIG. 1K1.
In another embodiment in the network 100B of
The switches, corresponding to the middle stages that are placed together are implemented as two four by four switches. For example middle switches MS(1,1) and MS(1,17) are placed together; so middle switch MS(1,1) is implemented as four by four switch with middle links ML(1,1), ML(1,7), ML(7,1) and ML(7,11) being the inputs and middle links ML(2,1), ML(2,3), ML(8,1) and ML(8,3) being the outputs; middle switch MS(1,17) is implemented as four by four switch with the middle links ML(1,2), ML(1,8), ML(7,2) and ML(7,12) being the inputs and middle links ML(2,2), ML(2,4), ML(8,2) and ML(8,4) being the outputs. Similarly in this embodiment of network 100B all the switches that are placed together are implemented as a two combined switches.
Layout diagrams 100C in
Referring to diagram 100L of
That is the switches that are placed together in Block 1_2 as shown in
Combined input and output switch IS1&OS1 is implemented as six by six switch with the inlet links IL1, IL2, ML(8,1), ML(8,2), ML(8,7) and ML(8,8) being the inputs and middle links ML(1,1)-ML(1,4) and outlet links OL1-OL2 being the outputs.
Middle switch MS(1,1) is implemented as four by four switch with middle links ML(1,1), ML(1,7), ML(7,1) and ML(7,11) being the inputs and middle links ML(2,1), ML(2,3), ML(8,1) and ML(8,3) being the outputs; And middle switch MS(1,17) is implemented as four by four switch with the middle links ML(1,2), ML(1,8), ML(7,2) and ML(7,12) being the inputs and middle links ML(2,2), ML(2,4), ML(8,2) and ML(8,4) being the outputs. Similarly all the other middle switches are also implemented as two four by four switches as illustrated in 100L of
In another embodiment, middle switch MS(1,1) (or the middle switches in any of the middle stage excepting the root middle stage) of Block 1_2 of Vbft(N1, N2, d, s) can be implemented as a two by four switch and a two by two switch to save cross points. This is because the left going middle links of these middle switches are never setup to the right going middle links. For example, in middle switch MS(1,1) of Block 1_2 as shown
In one embodiment, in the network 100B of
The switches, corresponding to the middle stages that are placed together are implemented as a four by four switch. For example middle switches MS(1,1) is implemented as four by four switch with middle links ML(1,1), ML(1,3), ML(7,1) and ML(7,5) being the inputs and middle links ML(2,1), ML(2,2), ML(8,1) and ML(8,2) being the outputs.
Layout diagrams 100C in
Referring to diagram 100L1 of FIG. 1L1 illustrates a high-level implementation of Block 1_2 (Each of the other blocks have similar implementation) for the layout 100C of
That is the switches that are placed together in Block 1_2 as shown in FIG. 1L1 are namely the input and output switch IS1&OS1 belonging to switch 1, illustrated by dotted lines, (as noted before switch 1 is for illustration purposes only, in practice the switches implemented are input switch IS1 and output switch OS1); middle switch MS(1,1) belonging to switch 2; middle switch MS(2,1) belonging to switch 3; middle switch MS(3,1) belonging to switch 4; And middle switch MS(4,1) belonging to switch 5.
Input and output switch IS1&OS1 are placed together; so input and output switch IS1&OS1 is implemented as four by four switch with the inlet links IL1, IL2, ML(8,1) and ML(8,3) being the inputs and middle links ML(1,1)-ML(1,2) and outlet links OL1-OL2 being the outputs.
Middle switch MS(1,1) is implemented as four by four switch with middle links ML(1,1), ML(1,3), ML(7,1) and ML(7,5) being the inputs and middle links ML(2,1), ML(2,2), ML(8,1) and ML(8,2) being the outputs. Similarly all the other middle switches are also implemented as four by four switches as illustrated in 100L1 of FIG. 1L1.
In another embodiment, middle switch MS(1,1) (or the middle switches in any of the middle stage excepting the root middle stage) of Block 1_2 of Vmlink-bft(N1, N2, d, s) can be implemented as a two by four switch and a two by two switch to save cross points. This is because the left going middle links of these middle switches are never setup to the right going middle links For example, in middle switch MS(1,1) of Block 1_2 as shown FIG. 1L1, the left going middle links namely ML(7,1) and ML(7,5) are never switched to the right going middle links ML(2,1) and ML(2,2). And hence to implement MS(1,1) two switches namely: 1) a two by four switch with the middle links ML(1,1) and ML(1,3) as inputs and the middle links ML(2,1), ML(2,2), ML(8,1), and ML(8,2) as outputs and 2) a two by two switch with the middle links ML(7,1) and ML(7,5) as inputs and the middle links ML(8,1) and ML(8,2) as outputs are sufficient without loosing any connectivity of the embodiment of MS(1,1) being implemented as an eight by eight switch as described before.)
Hypercube-Like Topology Layout Schemes:
Referring to diagram 300A in
As disclosed in PCT Application Serial No. PCT/US08/64604 that is incorporated by reference above, such a network can be operated in rearrangeably non-blocking manner for arbitrary fan-out multicast connections and also can be operated in strictly non-blocking manner for unicast connections.
The diagram 300A in
Each of the
middle switches are connected to exactly d switches in middle stage 160 through two links each for a total of 2×d links (for example the links ML(4,1) and ML(4,2) are connected from middle switch MS(3,1) to middle switch MS(4,1), and the links ML(4,3) and ML(4,4) are connected from middle switch MS(3,1) to middle switch MS(4,15)).
Each of the
middle switches MS(4,1)-MS(4,16) in the middle stage 160 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(4,1) and ML(4,2) are connected to the middle switch MS(4,1) from input switch MS(3,1), and the links ML(4,59) and ML(4,60) are connected to the middle switch MS(4,1) from input switch MS(3,15)) and also are connected to exactly d switches in middle stage 170 through two links each for a total of 2×d links (for example the links ML(5,1) and ML(5,2) are connected from middle switch MS(4,1) to middle switch MS(5,1), and the links ML(5,3) and ML(5,4) are connected from middle switch MS(4,1) to middle switch MS(5,15)).
Each of the
middle switches MS(5,1)-MS(5,16) in the middle stage 170 are connected from exactly d input switches through two links each for a total of 2×d links (for example the links ML(5,1) and ML(5,2) are connected to the middle switch MS(5,1) from input switch MS(4,1), and the links ML(5,59) and ML(5,60) are connected to the middle switch MS(5,1) from input switch MS(4,15)).
Finally the connection topology of the network 100A shown in
Referring to diagram 300B in
Just the same way there is difference in the connection topology between diagram 100A of
In one embodiment, in the network 300B of
Referring to layout 300C of
All the straight middle links are illustrated in layout 300C of
Even though it is not illustrated in layout 300C of
Recursively in each quadrant there are four sub-quadrants. For example in top-left quadrant there are four sub-quadrants namely top-left sub-quadrant, bottom-left sub-quadrant, top-right sub-quadrant and bottom-right sub-quadrant. Top-left sub-quadrant of top-left quadrant implements Block 1_2. Bottom-left sub-quadrant of top-left quadrant implements Block 3_4. Top-right sub-quadrant of top-left quadrant implements Block 7_8. Finally bottom-right sub-quadrant of top-left quadrant implements Block 5_6. Similarly there are two sub-halves in each quadrant. For example in top-left quadrant there are two sub-halves namely left-sub-half and right-sub-half. Left-sub-half of top-left quadrant implements Block 1_2 and Block 3_4. Right-sub-half of top-left quadrant implements Block 7_8 and Block 5_6. Recursively in larger multi-stage network Vfold-mlink(N1, N2, d, s) where N1=N2>32, the layout in this embodiment in accordance with the current invention, will be such that the super-quadrants will also be arranged in a similar manner.
Layout 300D of
Layout 300E of
Layout 300F of
Layout 300G of
The complete layout for the network 300B of
Some of the key aspects of the current invention related to layout diagram 300C of
The layout 300C in
Bottom-left super-quadrant implements the blocks from block 33_34 to block 63_64. Top-right super-quadrant implements the blocks from block 65_66 to block 95_96. And bottom-right super-quadrant implements the blocks from block 97_98 to block 127_128. In all these three super-quadrants also, the inter-block link connection topology is the exactly the same between the switches 1 and 2; switches 2 and 3; switches 3 and 4; switches 4 and 5 as that of the top-left super-quadrant.
Recursively in accordance with the current invention, the inter-block links connecting the switch 5 and switch 6 will be vertical tracks between the corresponding switches of top-left super-quadrant and bottom-left super-quadrant. And similarly the inter-block links connecting the switch 5 and switch 6 will be vertical tracks between the corresponding switches of top-right super-quadrant and bottom-right super-quadrant. The inter-block links connecting the switch 6 and switch 7 will be horizontal tracks between the corresponding switches of top-left super-quadrant and top-right super-quadrant. And similarly the inter-block links connecting the switch 6 and switch 7 will be horizontal tracks between the corresponding switches of bottom-left super-quadrant and bottom-right super-quadrant.
Ring Topology Layout Schemes:
Layout diagram 400C of
Referring to layout 400C of
All the straight middle links are illustrated in layout 400C of
Even though it is not illustrated in layout 400C of
Layout 400B of
Layout 400C of
Layout 400D of
Layout 400E of
The complete layout for the network 100B of
Some of the key aspects of the current invention related to layout diagram 400A of
Layout diagram 400C1 of FIG. 4C1 is another embodiment for the generalized folded multi-link multi-stage network Vfold-mlink(N1, N2, d, s) diagram 100B in
All the straight middle links are illustrated in layout 400C1 of FIG. 4C1. For example in Block 1_2, inlet links IL1-IL2, outlet links OL1-OL2, middle link ML(1,1), middle link ML(1,2), middle link ML(8,1), middle link ML(8,2), middle link ML(2,1), middle link ML(2,2), middle link ML(7,1), middle link ML(7,2), middle link ML(3,1), middle link ML(3,2), middle link ML(6,1), middle link ML(6,2), middle link ML(4,1), middle link ML(4,2), middle link ML(5,1) and middle link ML(5,2) are illustrated in layout 400C1 of FIG. 4C1.
Even though it is not illustrated in layout 400C1 of FIG. 4C1, in each block, in addition to the switches there may be Configurable Logic Blocks (CLB) or any arbitrary digital circuit or sub-integrated circuit block depending on the applications in different embodiments. The topology of the layout 400C1 in FIG. 4C1 is another embodiment of ring layout topology. For each of the neighboring rows in diagram 100B of
All the layout embodiments disclosed in the current invention are applicable to generalized multi-stage networks V(N1, N2, d, s), generalized folded multi-stage networks Vfold(N1, N2, d, s), generalized butterfly fat tree networks Vbft(N1, N2, d, s), generalized multi-link multi-stage networks Vmlink(N1, N2, d, s), generalized folded multi-link multi-stage networks Vfold-mlink(N1, N2, d, s), generalized multi-link butterfly fat tree networks Vmlink-bft(N1, N2, d, s), and generalized hypercube networks Vhcube(N1, N2, d, s) for s=1, 2, 3 or any number in general, and for both N1=N=N and N1≠N2, and d is any integer.
Conversely applicant makes another important observation that generalized hypercube networks Vhcube(N1, N2, d, s) are implemented with the layout topology being the hypercube topology shown in layout 100C of
All the embodiments disclosed in the current invention are useful in many varieties of applications. FIG. 5A1 illustrates the diagram of 500A1 which is a typical two by two switch with two inlet links namely IL1 and IL2, and two outlet links namely OL1 and OL2. The two by two switch also implements four crosspoints namely CP(1,1), CP(1,2), CP(2,1) and CP(2,2) as illustrated in FIG. 5A1. For example the diagram of 500A1 may the implementation of middle switch MS(1,1) of the diagram 100K of
All the embodiments disclosed in the current invention are useful in programmable integrated circuit applications. FIG. 5A2 illustrates the detailed diagram 500A2 for the implementation of the diagram 500A1 in programmable integrated circuit embodiments. Each crosspoint is implemented by a transistor coupled between the corresponding inlet link and outlet link, and a programmable cell in programmable integrated circuit embodiments. Specifically crosspoint CP(1,1) is implemented by transistor C(1,1) coupled between inlet link IL1 and outlet link OL1, and programmable cell P(1,1); crosspoint CP(1,2) is implemented by transistor C(1,2) coupled between inlet link IL1 and outlet link OL2, and programmable cell P(1,2); crosspoint CP(2,1) is implemented by transistor C(2,1) coupled between inlet link IL2 and outlet link OL1, and programmable cell P(2,1); and crosspoint CP(2,2) is implemented by transistor C(2,2) coupled between inlet link IL2 and outlet link OL2, and programmable cell P(2,2).
If the programmable cell is programmed ON, the corresponding transistor couples the corresponding inlet link and outlet link. If the programmable cell is programmed OFF, the corresponding inlet link and outlet link are not connected. For example if the programmable cell P(1,1) is programmed ON, the corresponding transistor C(1,1) couples the corresponding inlet link IL1 and outlet link OL1. If the programmable cell P(1,1) is programmed OFF, the corresponding inlet link IL1 and outlet link OL1 are not connected. In volatile programmable integrated circuit embodiments the programmable cell may be an SRAM (Static Random Address Memory) cell. In non-volatile programmable integrated circuit embodiments the programmable cell may be a Flash memory cell. Also the programmable integrated circuit embodiments may implement field programmable logic arrays (FPGA) devices, or programmable Logic devices (PLD), or Application Specific Integrated Circuits (ASIC) embedded with programmable logic circuits or 3D-FPGAs.
FIG. 5A2 also illustrates a buffer B1 on inlet link IL2. The signals driven along inlet link IL2 are amplified by buffer B1. Buffer B1 can be inverting or non-inverting buffer. Buffers such as B1 are used to amplify the signal in links which are usually long.
All the embodiments disclosed in the current invention are useful in one-time programmable integrated circuit applications. FIG. 5A3 illustrates the detailed diagram 500A3 for the implementation of the diagram 500A1 in one-time programmable integrated circuit embodiments. Each crosspoint is implemented by a via coupled between the corresponding inlet link and outlet link in one-time programmable integrated circuit embodiments. Specifically crosspoint CP(1,1) is implemented by via V(1,1) coupled between inlet link IL1 and outlet link OL1; crosspoint CP(1,2) is implemented by via V(1,2) coupled between inlet link IL1 and outlet link OL2; crosspoint CP(2,1) is implemented by via V(2,1) coupled between inlet link IL2 and outlet link OL1; and crosspoint CP(2,2) is implemented by via V(2,2) coupled between inlet link IL2 and outlet link OL2.
If the via is programmed ON, the corresponding inlet link and outlet link are permanently connected which is denoted by thick circle at the intersection of inlet link and outlet link. If the via is programmed OFF, the corresponding inlet link and outlet link are not connected which is denoted by the absence of thick circle at the intersection of inlet link and outlet link For example in the diagram 500A3 the via V(1,1) is programmed ON, and the corresponding inlet link IL1 and outlet link OL1 are connected as denoted by thick circle at the intersection of inlet link IL1 and outlet link OL1; the via V(2,2) is programmed ON, and the corresponding inlet link IL2 and outlet link OL2 are connected as denoted by thick circle at the intersection of inlet link IL2 and outlet link OL2; the via V(1,2) is programmed OFF, and the corresponding inlet link IL1 and outlet link OL2 are not connected as denoted by the absence of thick circle at the intersection of inlet link IL1 and outlet link OL2; the via V(2,1) is programmed OFF, and the corresponding inlet link IL2 and outlet link OL1 are not connected as denoted by the absence of thick circle at the intersection of inlet link IL2 and outlet link OL1. One-time programmable integrated circuit embodiments may be anti-fuse based programmable integrated circuit devices or mask programmable structured ASIC devices.
All the embodiments disclosed in the current invention are useful in Integrated Circuit Placement and Route applications, for example in ASIC backend Placement and Route tools. FIG. 5A4 illustrates the detailed diagram 500A4 for the implementation of the diagram 500A1 in Integrated Circuit Placement and Route embodiments. In an integrated circuit since the connections are known a-priori, the switch and crosspoints are actually virtual. However the concept of virtual switch and virtual crosspoint using the embodiments disclosed in the current invention reduces the number of required wires, wire length needed to connect the inputs and outputs of different netlists and the time required by the tool for placement and route of netlists in the integrated circuit.
Each virtual crosspoint is used to either to hardwire or provide no connectivity between the corresponding inlet link and outlet link. Specifically crosspoint CP(1,1) is implemented by direct connect point DCP(1,1) to hardwire (i.e., to permanently connect) inlet link IL1 and outlet link OL1 which is denoted by the thick circle at the intersection of inlet link IL1 and outlet link OL1; crosspoint CP(2,2) is implemented by direct connect point DCP(2,2) to hardwire inlet link IL2 and outlet link OL2 which is denoted by the thick circle at the intersection of inlet link IL2 and outlet link OL2. The diagram 500A4 does not show direct connect point DCP(1,2) and direct connect point DCP(1,3) since they are not needed and in the hardware implementation they are eliminated. Alternatively inlet link IL1 needs to be connected to outlet link OL1 and inlet link IL1 does not need to be connected to outlet link OL2. Also inlet link IL2 needs to be connected to outlet link OL2 and inlet link IL2 does not need to be connected to outlet link OL1. Furthermore in the example of the diagram 500A4, there is no need to drive the signal of inlet link IL1 horizontally beyond outlet link OL1 and hence the inlet link IL1 is not even extended horizontally until the outlet link OL2. Also the absence of direct connect point DCP(2,1) illustrates there is no need to connect inlet link IL2 and outlet link OL1.
In summary in integrated circuit placement and route tools, the concept of virtual switches and virtual cross points is used during the implementation of the placement & routing algorithmically in software, however during the hardware implementation cross points in the cross state are implemented as hardwired connections between the corresponding inlet link and outlet link, and in the bar state are implemented as no connection between inlet link and outlet link.
All the embodiments disclosed in the current invention are also useful in the design of SoC interconnects, Field programmable interconnect chips, parallel computer systems and in time-space-time switches.
Numerous modifications and adaptations of the embodiments, implementations, and examples described herein will be apparent to the skilled artisan in view of the disclosure.
This application is related to and claims priority of the PCT Application Serial No. PCT/US08/64605 entitled “VLSI LAYOUTS OF FULLY CONNECTED GENERALIZED NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 22, 2008, and the U.S. Provisional Patent Application Ser. No. 60/940,394 entitled “VLSI LAYOUTS OF FULLY CONNECTED GENERALIZED NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 25, 2007. This application is related to and incorporates by reference in its entirety the U.S. application Ser. No. 12/530,207 entitled “FULLY CONNECTED GENERALIZED MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed Sep. 6, 2009, the PCT Application Serial No. PCT/US08/56064 entitled “FULLY CONNECTED GENERALIZED MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed Mar. 6, 2008, the U.S. Provisional Patent Application Ser. No. 60/905,526 entitled “LARGE SCALE CROSSPOINT REDUCTION WITH NONBLOCKING UNICAST & MULTICAST IN ARBITRARILY LARGE MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed Mar. 6, 2007, and the U.S. Provisional Patent Application Ser. No. 60/940,383 entitled “FULLY CONNECTED GENERALIZED MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 25, 2007. This application is related to and incorporates by reference in its entirety the U.S. patent application Ser. No. 12/601,273 entitled “FULLY CONNECTED GENERALIZED BUTTERFLY FAT TREE NETWORKS” by Venkat Konda assigned to the same assignee as the current application filed concurrently, the PCT Application Serial No. PCT/US08/64603 entitled “FULLY CONNECTED GENERALIZED BUTTERFLY FAT TREE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 22, 2008, the U.S. Provisional Patent Application Ser. No. 60/940, 387 entitled “FULLY CONNECTED GENERALIZED BUTTERFLY FAT TREE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 25, 2007, and the U.S. Provisional Patent Application Ser. No. 60/940, 390 entitled “FULLY CONNECTED GENERALIZED MULTI-LINK BUTTERFLY FAT TREE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 25, 2007. This application is related to and incorporates by reference in its entirety the U.S. patent application Ser. No. 12/601,274 entitled “FULLY CONNECTED GENERALIZED MULTI-LINK MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application filed concurrently, the PCT Application Serial No. PCT/US08/64604 entitled “FULLY CONNECTED GENERALIZED MULTI-LINK MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 22, 2008, the U.S. Provisional Patent Application Ser. No. 60/940, 389 entitled “FULLY CONNECTED GENERALIZED REARRANGEABLY NONBLOCKING MULTI-LINK MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 25, 2007, the U.S. Provisional Patent Application Ser. No. 60/940, 391 entitled “FULLY CONNECTED GENERALIZED FOLDED MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 25, 2007 and the U.S. Provisional Patent Application Ser. No. 60/940,392 entitled “FULLY CONNECTED GENERALIZED STRICTLY NONBLOCKING MULTI-LINK MULTI-STAGE NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed May 25, 2007. This application is related to and incorporates by reference in its entirety the U.S. Provisional Patent Application Ser. No. 61/252, 603 entitled “VLSI LAYOUTS OF FULLY CONNECTED NETWORKS WITH LOCALITY EXPLOITATION” by Venkat Konda assigned to the same assignee as the current application, filed Oct. 16, 2009. This application is related to and incorporates by reference in its entirety the U.S. Provisional Patent Application Ser. No. 61/252, 609 entitled “VLSI LAYOUTS OF FULLY CONNECTED GENERALIZED AND PYRAMID NETWORKS” by Venkat Konda assigned to the same assignee as the current application, filed Oct. 16, 2009.
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/US2008/064605 | 5/22/2008 | WO | 00 | 5/31/2010 |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO2008/147928 | 12/4/2008 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 4813038 | Lee | Mar 1989 | A |
| 5406556 | Widjaja et al. | Apr 1995 | A |
| 5856977 | Yang et al. | Jan 1999 | A |
| 6049542 | Prasad | Apr 2000 | A |
| 6157643 | Ma | Dec 2000 | A |
| 6567858 | Yang et al. | May 2003 | B1 |
| 6940308 | Wong | Sep 2005 | B2 |
| 7130920 | Sailor | Oct 2006 | B1 |
| 7136380 | Li | Nov 2006 | B2 |
| 7154887 | Wu et al. | Dec 2006 | B2 |
| 7346049 | Towles | Mar 2008 | B2 |
| 7349387 | Wu | Mar 2008 | B2 |
| 7397796 | Smiljani | Jul 2008 | B1 |
| 7424010 | Konda | Sep 2008 | B2 |
| 7424011 | Konda | Sep 2008 | B2 |
| 7468974 | Carr et al. | Dec 2008 | B1 |
| 7924052 | Feng et al. | Apr 2011 | B1 |
| 8098081 | Trimberger | Jan 2012 | B1 |
| Number | Date | Country | |
|---|---|---|---|
| 20110037498 A1 | Feb 2011 | US |
| Number | Date | Country | |
|---|---|---|---|
| 60940394 | May 2007 | US |
