Heterogeneous Memory Design Exploration for AI Accelerators with a Gain Cell Memory Compiler

Feb 24, 2026 by Xinxin Wang, Lixian Yan, Shuhan Liu, Luke R. Upton, Zhuoqi Cai, Yiming Tan, Shengman Li, K. Jana, Peijing Li, Jesse Cirimelli-Low, Thierry Tambe, M. Guthaus, H.-S. Philip Wong

We built OpenGCRAM, a compiler that spits out layout-ready SRAM and Gain Cell RAM macros for commercial CMOS and profiles area, delay, and power so you can actually explore heterogeneous on-chip memories for AI accelerators. If you care about squeezing density and energy by mixing tunable-retention GCRAM with SRAM, this tool lets you systematically find the sweet spots instead of guessing.

source S2