Pan Zhao

Pan Zhao

Auto-generated profile, based on publicly available information

Pan Zhao

Shanghai Jiao Tong University, Shanghai, Shanghai, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 March 2025 in IEEE Transactions on Circuits and Systems II: Express Briefs
Verified
Journal Article
Closed access
PZPan Zhao
Pan Zhao
DXDonghui Xue
Donghui Xue
LWLicheng Wu
Licheng Wu
Liang ChangLiang Chang
Liang Chang
HTHaining Tan
Haining Tan
Jun ZhouJun Zhou
Jun Zhou
Quantization is an important technique for the acceleration of transformer-based neural networks. Prior related works mainly consider quantization from the algorithm level. Their hardware implementation is inefficient. In this work, we propose an efficient vision transformer accelerator with retraining-free and finetuning-free hybrid-precision quantization. At the algorithm level, the features and weights are divided into two parts: normal values and outlier values. These two parts are quantized with different bit widths and scaling factors. We use matrix transformation and group-wise quantization policy to improve hardware utilization. At the hardware level, we propose a two-stage FIFO group structure and a hierarchical interleaving data flow to further improve the utilization of the PE array. As a result, the input and weight matrices are quantized to 5.71 bits on average with 0.526% accuracy loss on Swin-T. The accelerator achieves a frame rate of 118.9 FPS and an energy efficiency of 43.58 GOPS/W on the ZCU102 FPGA board, better than state-of-the-art works.

1 October 2024 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Verified
Journal Article
Closed access
YWYuanbo Wang
Yuanbo Wang
Liang ChangLiang Chang
Liang Chang
JWJingke Wang
Jingke Wang
PZPan Zhao
Pan Zhao
JZJiahao Zeng
Jiahao Zeng
Jun ZhouJun Zhou
Jun Zhou
Computing-in-memory (CIM) architecture has become a promising solution to improve the parallelism of the multiply-and-accumulation (MAC) operation for artificial intelligence (AI) processors. Recently, revived CIM engine partly relieves the memory wall issue by integrating computation in/with the memory. However, current CIM solutions still require large data movements with the increase of the practical neural network model and massive input data. Previous CIM works only considered computation without concern for the memory attribute, leading to a low memory computing ratio. This article presents a static-random access-memory (SRAM)-based digital CIM macro supporting pipeline mode and computation-memory-aware technique to improve the memory computing ratio. We develop a novel weight driver with fine-grained ping-pong operation, avoiding the computation stall caused by weight update. Based on our evaluation, the peak energy efficiency is 19.78 TOPS/W at the 22-nm technology node, 8-bit width, and 50% sparsity of the input feature map.

6 November 2023 in IEEE Transactions on Circuits and Systems II: Express Briefs
Verified
Journal Article
Closed access
PZPan Zhao
Pan Zhao
Liang ChangLiang Chang
Liang Chang
JZJiahao Zeng
Jiahao Zeng
LWLicheng Wu
Licheng Wu
LZLiang Zhou
Liang Zhou
Jun ZhouJun Zhou
Jun Zhou
Voxel-based neural networks are promising algorithms to process 3D point cloud data. However, the voxel-based neural networks run inefficiently on GPUs for the unique calculation modes of 3D sparse convolution. In addition, the large irregular activation sparsity of 3D voxel-based neural networks can be utilized more to improve the performance of the hardware accelerator. This work presents a high-performance 3D point cloud accelerator, namely HISPOC, which can effectively leverage irregular activation sparsity. Specifically, we propose a 3D bitmap-based search engine that significantly reduces the complexity of voxel searching. In addition, we propose a match-based input feature manager and a dichotomy-based output feature sparsity encoder to enhance computation efficiency and reduce RAM consumption based on the Zero-Value Compression algorithm. Finally, the average memory footprint of the feature is reduced to 54.98%. we implement the HISPOC on Xilinx Virtex VC707 board, achieving 3.3-1272.1 GOPS, and 11.8 FPS under 109.8 KB RAM consumption, which are 3.34×, 1.39×, and 0.62× compared with the state-of-the-art works.

4 December 2025 in IEEE Transactions on Circuits and Systems I: Regular Papers
Verified
Journal Article
Closed access
PZPan Zhao
Pan Zhao
LWLicheng Wu
Licheng Wu
TYTing Yue
Ting Yue
XZxin zhao
xin zhao
JZJing Zhou
Jing Zhou
Liang ChangLiang Chang
Liang Chang
Voxel-based neural networks have been widely applied to the processing of large-scale outdoor point cloud data, which first convert points into voxels and then extract features using several sparse convolution and normal convolution layers. The hardware implementation of these networks suffers from complex rulebook generation, irregular memory access, and low hardware utilization. Meanwhile, these networks still have much data sparsity. In this paper, we propose an energy and area-efficient query-based sparsity-aware accelerator for voxel-based point cloud neural networks, namely QSAP. Specifically, a dedicated unit is used to improve the efficiency of rulebook generation. A query-based input feature-writing method is proposed to enhance parallel reading potential. An efficient weight-mapping method is introduced to store unpruned weights in on-chip buffers. A novel input-feature-reading method with consecutive queries is proposed to enable out-of-order execution of convolution operations, thereby improving hardware utilization. The hardware utilization is further enhanced by a proposed pop strategy that minimizes the total number of empty FIFOs. The MAC related to zero value is also skipped in this process. As a result, QSAP achieves superior performance on 22 nm technology with the throughput, energy efficiency, area efficiency, frame rate, and frame energy of 1074 GOPS, 7.79 TOPS/W, 590 GOPS/mm $\mathbf {^{2}}$ , 35.2 FPS, and 3.91 mJ/Frame, respectively, better than state-of-the-art works.

26 July 2026
Verified
Conference Paper
Open access
RJRenda Jian
Renda Jian
YCYuzhou Chen
Yuzhou Chen
PZPan Zhao
Pan Zhao
YCYitian Chen
Yitian Chen
ZWZhican Wang
Zhican Wang
GHGuanghui He
Guanghui He

26 July 2026
Verified
Conference Paper
Open access
XWXueling Wang
Xueling Wang
YCYuzhou Chen
Yuzhou Chen
ZWZhican Wang
Zhican Wang
PZPan Zhao
Pan Zhao
RJRenda Jian
Renda Jian
GHGuanghui He
Guanghui He