Why another Routing Fabric? Why should FPGA vendors adopt it?

  • What are the issues with prevailing FPGA fabrics?
    • Prevailing FPGA devices have forever been extremely¬†area inefficient. They are based on 2D-Mesh Network with N^2 crosspoint complexity.
    • 2D-Mesh based FPGAs use ~75% die area for fabric.
  • Is there an alternative fabric?
    • Multi-stage based networks with N*log N crosspoint complexity are known since the invention of automatic telephone switching.
    • However, so far, no practical layouts for Multi-stage networks are known to be usable as FPGA Fabrics.
  • How hard is it to sell Multi-stage networks as FPGA Fabric?
    • Since the invention of FPGA, no one found out viable and elegant layout for multi-stage network based fabric that compete with 2D-Mesh Fabrics in every dimension.
    • FPGA Fabric architects/designers, in general, have no understanding of multi-stage networks
      • Previous failed multi-stage Fabric attempts did not solve the problem as elegantly in every dimension.
      • Furthermore previous failed attempts to use multi-stage networks made FPGA Fabric designers even more suspicious of them.