Skip to content
Tile based architecture designed for computing efficiency, scalability and generality
Branch: master
Clone or download


This repo contains the bsg_manycore source code designed by Bespoke Silicon Group@University of Washington.

The tile based architecture is designed for computing efficiency, scalability and generality. The two main components are:

  • Computing Node: In-house RISC-V 32IM compatible core runs at 1.4GHz@16nm, but nodes also can be any other accelerators.
  • Mesh Network : Dimension ordered, single flit network with inter-nodes synchronization primitives (mutex, barrier etc.)

Without any customized circuit, a 16nm prototype chip that holds 16x31 tiles on a 4.5x3.4 mm^2 die space achieves 812,350 aggregated CoreMark score.


  1. Chip gallery, publications, and artworks:
  2. Bleeding edge features and proceedings:


Comming Soon!

You can’t perform that action at this time.