A Survey of Machine Learning for Computer Architecture and Systems

Abstract

It has been a long time that computer architecture and systems are optimized to enable efficient execution of machine learning (ML) algorithms or models. Now, it is time to reconsider the relationship between ML and systems, and let ML transform the way that computer architecture and systems are designed. This embraces a twofold meaning: the improvement of designers' productivity, and the completion of the virtuous cycle. In this paper, we present a comprehensive review of work that applies ML for system design, which can be grouped into two major categories, ML-based modelling that involves predictions of performance metrics or some other criteria of interest, and ML-based design methodology that directly leverages ML as the design tool. For ML-based modelling, we discuss existing studies based on their target level of system, ranging from the circuit level to the architecture/system level. For ML-based design methodology, we follow a bottom-up path to review current work, with a scope of (micro-)architecture design (memory, branch prediction, NoC), coordination between architecture/system and workload (resource allocation and management, data center management, and security), compiler, and design automation. We further provide a future vision of opportunities and potential directions, and envision that applying ML for computer architecture and systems would thrive in the community. CCS Concepts: • Computing methodologies → Machine learning; • Computer systems organization → Architectures; • General and reference → Surveys and overviews.

Hybrid Graph Models for Logic Optimization via Spatio-Temporal Information

“…Multi-modal graph representation learning. Prior arts that adopt GNNs for fast evaluation focus on mapping static circuit graphs to metric of interest [12,13]. The most significant innovation of this work is to consider input information from multiple modality, i.e., circuit designs in graph format and synthesis flows in sequence format.…”

“…From a broader view, GNNs are expected to better utilize graph structured data in many EDA problems. Instead of conventional graph representation learning that maps circuit designs from static graphs to labels (e.g., resource/timing/power) [12,13], the target task in this work should consider both designs (i.e., static graphs) and synthesis flows (i.e., transformations to be applied on graphs) to provide high-accuracy predictions of delay/area, which can be recognized as a multimodal or dynamic graph representation learning.…”

“…Recent years have witnessed the naissance of machine The proposed model exploits spatial information from circuit designs and temporal knowledge from synthesis flows, generalizable to new designs without invoking any retraining. [13], where one promising direction is ML-enabled fast performance modeling. Industrial investigations [4] highlighted two basic requirements for production-ready ML in EDA: 1 ○ the accuracy of ML-based performance estimation should be a minimum of 2𝜎 (∼95%); 2 ○ generalization capability is important for production-ready ML applications, meaning that an ML algorithm should be able to directly apply to new designs without retraining.…”

High-Level Synthesis Performance Prediction using GNNs: Benchmarking, Modeling, and Advancing

“…Analytical models are classic approaches [19,32,33] but they only work for highly regular dataflow such as perfect loops and arrays. Recent ML approaches have become promising in estimating the actual design performance [29]. Pyramid [15] assembled multiple ML models for resource and timing prediction.…”