OpenACMv2: An Accuracy-Constrained Co-Optimization Framework for Approximate DCiM

2026-03-13

AI Summary

This work addresses the challenge of balancing energy efficiency and computational accuracy in approximate compute-in-memory (Approximate DCiM) systems by proposing OpenACMv2, a precision-constrained co-optimization framework. OpenACMv2 innovatively decouples the optimization into two stages: the first employs a graph neural network (GNN)-based surrogate model to perform architecture search under explicit accuracy constraints, while the second leverages Monte Carlo methods to optimize transistor sizing with awareness of process, voltage, temperature variations, and manufacturing-induced process corners. By integrating single-objective and multi-objective optimization strategies, the framework ensures convergence while achieving Pareto-optimal trade-offs among power, performance, area (PPA), and accuracy. Evaluated on the FreePDK45 technology node using the OpenROAD toolchain, OpenACMv2 demonstrates significant PPA improvements under controllable accuracy loss and enables rapid “what-if” design exploration. The implementation is publicly released.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Optimization
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Digital Compute-in-Memory (DCiM) accelerates neural networks by reducing data movement. Approximate DCiM can further improve power-performance-area (PPA), but demands accuracy-constrained co-optimization across coupled architecture and transistor-level choices. Building on OpenYield, we introduce Accuracy-Constrained Co-Optimization (ACCO) and present OpenACMv2, an open framework that operationalizes ACCO via two-level optimization: (1) accuracy-constrained architecture search of compressor combinations and SRAM macro parameters, driven by a fast GNN-based surrogate for PPA and error; and (2) variation- and PVT-aware transistor sizing for standard cells and SRAM bitcells using Monte Carlo. By decoupling ACCO into architecture-level exploration and circuit-level sizing, OpenACMv2 integrates classic single- and multi-objective optimizers to deliver strong PPA-accuracy tradeoffs and robust convergence. The workflow is compatible with FreePDK45 and OpenROAD, supporting reproducible evaluation and easy adoption. Experiments demonstrate significant PPA improvements under controlled accuracy budgets, enabling rapid "what-if" exploration for approximate DCiM. The framework is available on https://github.com/ShenShan123/OpenACM.

Problem

• Approximate DCiM
• accuracy-constrained co-optimization
• PPA
• architecture-circuit co-design
• neural network acceleration

Innovation

• Approximate DCiM
• Accuracy-Constrained Co-Optimization
• GNN-based surrogate
• PVT-aware transistor sizing
• Open-source EDA