Wave-pipelining project "Shaping The Waves"
This project follows the idea of System Hyper Pipelining (SHP) and combines it with wave-pipelining.

SHP generates interleaved-multithreaded designs and replaces original registers with memories to allow dynamic interleaved-multithreaded scheduling (threads can be stalled and therefore reordered).

SHP also uses C-Slow Retiming. But instead of inserting "C-1" registers into each path, wave-pipelining is used in this project.

wp

Fig. 1. Wave-pipelining, all paths in a design must have equal length

Wave-pipelining requires equal length between the relevant registers. Multiple waves travel through the logic during execution. More information on this rather old idea of wave-pipelining can be found in this paper [1].

[1] W. Burleson, M. Ciesielski, F. Klass, and W. Liu, "Wave-Pipelining: A Tutorial and Research Survey", IEEE Trans. on VLSI, vol. 6, no. 3, Sept. 1998, link: https://www.cs.princeton.edu/courses/archive/fall01/cs597a/wave.pdf

The project is very similar to the Arduissimo project, which follows the same idea but is optimized for FPGA technology. The ASIC version now uses wave-pipelining instead of register insertion and has only one RISC-V core implemented (instead of 4).

The ASIC project "Shaping The Waves" is defined by:
  • System-Hyper-Pipelined MiniSoC
  • 232 MHz running 3 waves 8 independent threads
  • RV32iMC-P3C4D8W3
    • 3 classical Pipeline stages
    • 4 Copies running at the same time
    • memory thread Depth is 8
    • 3 Waves maximum
  • Timer with 128 programmable events
  • 32 GPIOs with programmable trigger-units for SW-defined protocols
  • SkyWater 0,13µm, Efabless Shuttle 8
  • OpenSource tools: Caravel\OpenLane
  • Sponsored by Efabless and Google !!!

The project can be found on github here.

The following Figure shows the ASIC implementation:

STW

Fig. 2. ASIC layout of the project
last modified: 2023/Mar/22