Jiaqi Gu

Jiaqi Gu

Assistant Professor in School of Electrical, Computer and Energy Engineering, Arizona State University

Softwares

GPU-Accelerated VLSI Placement

Published:

We release DREAMPlace 3.0, a Deep learning toolkit-enabled VLSI placement engine. DREAMPlace 3.0 supports region constraints with multi-electrostatic fields and enhanced gradient descent optimization with quadratic density penalty and adaptive entropy injection. With the analogy between nonlinear VLSI placement and deep learning training problem, this tool is developed with deep learning toolkit for flexibility and efficiency. The tool runs on both CPU and GPU. Over 30X speedup over the CPU implementation (RePlAce) is achieved in global placement and legalization on ISPD 2005 contest benchmarks with a Nvidia Tesla V100 GPU. DREAMPlace also integrates a GPU-accelerated detailed placer, ABCDPlace, which can achieve around 16X speedup on million-size benchmarks over the widely-adopted sequential placer NTUPlace3 on CPU.

Open-source PyTorch-centric Optical Neural Network Library

Published:

We release TorchONN, an open-source PyTorch-centric optical neural network library. We support fast and scalable development, training, and optimization of customized ONN layers and models. The tool runs on both CPU and GPU. CUDA-accelerated batched operators achieves 10-50X speedup over CPU-based matrix decomposition and unitary group parametrization.

PyTorch Training Utility

Published:

We release pyutils, a collection of pytorch model training utilities. We support general PyTorch training facilities and special computing operators used in pytorch-onn. Hierarchical configuration, various optimizers, lr_schedulers, efficient optical neurocomputing operators are supported.

Open-source Hybrid Classical-Quantum Machine Learning Framework

Published:

We release TorchQuantum, an open-source PyTorch-centric hybrid classical-quantum machine learning framework. We support easy construction of parameterized quantum circuits in PyTorch; bach mode inference and training on CPU/GPU;dynamic computation graph for easy debugging; and easy deployment on real quantum devices such as IBMQ. We can support more than 26-qubit simulation with order-of-magnitude faster speed than PennyLane. The official website is at qmlsys.