Bluespec Adds Custom RISC-V Instructions

Bluespec’s licensable RISC-V MCUX CPU IP adds custom instructions to its existing ultralow-footprint CPU, branded MCU. Optimized for FPGAs, these CPUs occupy fewer lookup tables than any other commercial RISC-V core.

Bryon Moyer

Bluespec has added a custom-instruction capability to its licensable ultralow-footprint RISC-V core for customers implementing a CPU in FPGA logic. The new MCUX core extends the feature set of the company’s MCU-branded CPU launched 3Q21. Both cores require 90% fewer hardware resources than the company’s smallest five-stage single-core model while raising the maximum clock speed.

MCU’s single-stage pipelined (or nonpipelined) microarchitecture, which Bluespec calls multicycle, serves applications requiring a small logic footprint and which can tolerate lower performance. Custom instructions will add to the logic footprint. The company provides templates for implementing the CPUs.

Bluespec began in 2003 as an EDA company offering a hardware description language. Although the company still maintains that language for internal use, it pivoted to offering RISC-V cores in 2014. Arvind Mithal cofounded the company on the basis of research at MIT, where he still teaches. Charlie Hauck leads the company as CEO, having previously worked at Faraday Technology, Lexra, and LSI Logic. CTO Rishiyur Nikhil hails out of Cambridge Research Labs and MIT. North Bridge Ventures is the primary investor; Bluespec completed a $5 million investment round last February.

Typical licensing, which is available now, involves an upfront fee and yearly maintenance; the company charges no royalties. Delivery includes RTL for integration and an AMD (Xilinx) bitstream for evaluation and as a reference implementation. The base CPU occupies in the range of 1,600 (for the MCU Lite version) to 2,200 (for MCUX) six-input lookup tables (LUTs) in an FPGA (excluding logic necessary for custom instructions). Clock frequency varies by FPGA but typically ranges from 100 to 330 MHz; benchmark numbers are unavailable. The current offering focuses on AMD FPGAs.