An Open-Source ML-Based Full-Stack Optimization Framework for Machine Learning Accelerators

Parameterizable machine learning (ML) accelerators are the product of recent breakthroughs in ML. To fully enable their design space exploration (DSE), we propose a physical-design-driven, learning-based prediction framework for hardware-accelerated deep neural network (DNN) and non-DNN ML algorithms. It adopts a unified approach that combines backend power, performance, and area (PPA) analysis with frontend performance simulation, thereby achieving a realistic estimation of both backend PPA and system metrics such as runtime and energy. In addition, our framework includes a fully automated DSE technique, which optimizes backend and system metrics through an automated search of architectural and backend parameters. Experimental studies show that our approach consistently predicts backend PPA and system metrics with an average 7% or less prediction error for the ASIC implementation of two deep learning accelerator platforms, VTA and VeriGOOD-ML, in both a commercial 12 nm process and a research-oriented 45 nm process.

Performance Analysis of DNN Inference/Training with Convolution and non-Convolution Operations

Today's performance analysis frameworks for deep learning accelerators suffer from two significant limitations. First, although modern convolutional neural network (CNNs) consist of many types of layers other than convolution, especially during training, these frameworks largely focus on convolution layers only. Second, these frameworks are generally targeted towards inference, and lack support for training operations. This work proposes a novel performance analysis framework, SimDIT, for general ASIC-based systolic hardware accelerator platforms. The modeling effort of SimDIT comprehensively covers convolution and non-convolution operations of both CNN inference and training on a highly parameterizable hardware substrate. SimDIT is integrated with a backend silicon implementation flow and provides detailed end-to-end performance statistics (i.e., data access cost, cycle counts, energy, and power) for executing CNN inference and training workloads. SimDIT-enabled performance analysis reveals that on a 64X64 processing array, non-convolution operations constitute 59.5% of total runtime for ResNet-50 training workload. In addition, by optimally distributing available off-chip DRAM bandwidth and on-chip SRAM resources, SimDIT achieves 18X performance improvement over a generic static resource allocation for ResNet-50 inference.

K-SpecPart: A Supervised Spectral Framework for Multi-Way Hypergraph Partitioning Solution Improvement

State-of-the-art hypergraph partitioners follow the multilevel paradigm, constructing multiple levels of coarser hypergraphs to drive cutsize refinement. These partitioners face limitations: (i) coarsening processes depend on local neighborhood structure, ignoring global hypergraph structure; (ii) refinement heuristics risk entrapment in local minima. We introduce K-SpecPart, a supervised spectral framework addressing these limitations by solving a generalized eigenvalue problem, capturing balanced partitioning objectives and global hypergraph structure in a low-dimensional vertex embedding while leveraging high-quality multilevel partitioning solutions as hints. In multi-way partitioning, K-SpecPart derives multiple bipartitioning solutions from a multi-way hint partitioning solution. It integrates these solutions into the generalized eigenvalue problem to compute eigenvectors, creating a large-dimensional embedding. Linear Discriminant Analysis (LDA) is used to transform this into a lower-dimensional embedding. K-SpecPart constructs a family of trees from the vertex embedding and partitions them using a tree-sweeping algorithm. We extend SpecPart's tree partitioning algorithm for multi-way partitioning. The multiple tree-based partitioning solutions are overlaid, followed by lifting to a clustered hypergraph where an integer linear programming (ILP) partitioning problem is solved. Empirical studies show K-SpecPart's benefits. For bipartitioning, K-SpecPart outperforms SpecPart with improvements up to 30%. For multi-way partitioning, K-SpecPart surpasses hMETIS and KaHyPar, with improvements up to 20% in some cases.

Assessment of Reinforcement Learning for Macro Placement

We provide open, transparent implementation and assessment of Google Brain's deep reinforcement learning approach to macro placement and its Circuit Training (CT) implementation in GitHub. We implement in open source key "blackbox" elements of CT, and clarify discrepancies between CT and Nature paper. New testcases on open enablements are developed and released. We assess CT alongside multiple alternative macro placers, with all evaluation flows and related scripts public in GitHub. Our experiments also encompass academic mixed-size placement benchmarks, as well as ablation and stability studies. We comment on the impact of Nature and CT, as well as directions for future research.