GENIAL: Generative Design Space Exploration via Network Inversion for Low Power Algorithmic Logic Units

TL;DR

GENIAL tackles the energy bottleneck in AI arithmetic units by automatically discovering low-power operand encodings through a Transformer-based surrogate trained in two stages and sharpened by network inversion. It embeds a hierarchical EDA-in-the-loop workflow with ESPRS, OPENR, and FLOWY to evaluate designs against power, area, and switching activity, guided by a learned QoR predictor built on a PointNet+Transformer backbone. The approach achieves notable gains, including encodings with up to $18\%$ lower power than the conventional two's complement in 4-bit multipliers, and demonstrates promising applicability to finite-state machines, while releasing the full toolkit for reproducibility. These results indicate a practical path toward quality-of-results-optimized combinational circuit generation through encoding-aware design space exploration.

Abstract

As AI workloads proliferate, optimizing arithmetic units is becoming increasingly important for reducing the footprint of digital systems. Conventional design flows, which often rely on manual or heuristic-based optimization, are limited in their ability to thoroughly explore the vast design space. In this paper, we introduce GENIAL, a machine learning-based framework for the automatic generation and optimization of arithmetic units, with a focus on multipliers. At the core of GENIAL is a Transformer-based surrogate model trained in two stages, involving self-supervised pretraining followed by supervised finetuning, to robustly forecast key hardware metrics such as power and area from abstracted design representations. By inverting the surrogate model, GENIAL efficiently searches for new operand encodings that directly minimize power consumption in arithmetic units for specific input data distributions. Extensive experiments on large datasets demonstrate that GENIAL is consistently more sample efficient than other methods, and converges faster towards optimized designs. This enables deployment of a high-effort logic synthesis optimization flow in the loop, improving the accuracy of the surrogate model. Notably, GENIAL automatically discovers encodings that achieve up to 18% switching activity savings within multipliers on representative AI workloads compared with the conventional two's complement. We also demonstrate the versatility of our approach by achieving significant improvements on Finite State Machines, highlighting GENIAL's applicability for a wide spectrum of logic functions. Together, these advances mark a significant step toward automated Quality-of-Results-optimized combinational circuit generation for digital systems.

Figures (14)

• Figure 1: GENIAL framework overview: Generator → EDA Flows → Surrogate Model → Recommender
• Figure 2: Representation of notable encodings as 2D tensors
• Figure 3: Custom high-effort logic synthesis optimization flow
• Figure 4: swact model for a cell with a fanout of two
• Figure 5: qmp model architecture