Haotian Fu

Haotian Fu

Auto-generated profile, based on publicly available information

Haotian Fu

MICS Thrust, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

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

Representative Publications

View All Publications
31 March 2025 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Verified
Journal Article
Closed access
HRHongwei Ren
Hongwei Ren
YZYue Zhou
Yue Zhou
JZJiadong Zhu
Jiadong Zhu
XLXiaopeng Lin
Xiaopeng Lin
HFHaotian Fu
Haotian Fu
BCBojun Cheng
Bojun Cheng
Event cameras draw inspiration from biological systems, boasting low latency and high dynamic range while consuming minimal power. The most current approach to processing Event Cloud often involves converting it into frame-based representations, which neglects the sparsity of events, loses fine-grained temporal information, and increases the computational burden. In contrast, Point Cloud is a popular representation for processing 3-dimensional data and serves as an alternative method to exploit local and global spatial features. Nevertheless, previous point-based methods show an unsatisfactory performance compared to the frame-based method in dealing with spatio-temporal event streams. In order to bridge the gap, we propose EventMamba, an efficient and effective framework based on Point Cloud representation by rethinking the distinction between Event Cloud and Point Cloud, emphasizing vital temporal information. The Event Cloud is subsequently fed into a hierarchical structure with staged modules to process both implicit and explicit temporal features. Specifically, we redesign the global extractor to enhance explicit temporal extraction among a long sequence of events with temporal aggregation and State Space Model (SSM) based Mamba. Our model consumes minimal computational resources in the experiments and still exhibits SOTA point-based performance on six different scales of action recognition datasets. It even outperformed all frame-based methods on both Camera Pose Relocalization (CPR) and eye-tracking regression tasks.

20 September 2021 in IEEE Transactions on Circuits and Systems I: Regular Papers
Verified
Journal Article
Closed access
HFHaotian Fu
Haotian Fu
QHQinghui Hong
Qinghui Hong
Chunhua WangChunhua Wang
Chunhua Wang
JSJingru Sun
Jingru Sun
YLYa Li
Ya Li
Inhomogeneous linear ordinary differential equations (ODEs) and systems of ODEs can be solved in a variety of ways. However, hardware circuits that can perform the efficient analog computation to solve them are rarely in the literature. To address such problems, this paper proposes a general method of using a memristor-capacitor (M-C) circuit to solve inhomogeneous linear ODEs and systems of ODEs of any order in initial value problems. The M-C circuit can match the coefficients of the equations sought by adjusting the memristor resistance value according to the coefficient formula proposed in the paper, which has higher programmability. Then, some ODEs and systems of ODEs are given in the paper as examples to evaluate the proposed method. According to the comparison results based on MATLAB software simulation and the simulation based on OrCAD software, the designed M-C circuit has an effective improvement in speed and the accuracy exceeds 99.95% in software simulation. Based on practical verification, this paper gives the actual M-C circuit experiment based on PCB. Moreover, the proposed method can be used to quickly solve the object motion state in the spring mass damping system in actual engineering, and the accuracy can reach 99.98%.

7 February 2023 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
Verified
Journal Article
Closed access
QHQinghui Hong
Qinghui Hong
HFHaotian Fu
Haotian Fu
YLYiyang Liu
Yiyang Liu
JZJiliang Zhang
Jiliang Zhang
Complex-valued neural networks have better optimization capabilities, stronger robustness and richer characterization capabilities compared with real-valued neural networks, which has achieved good results in the field of portrait restoration. However, there is almost no circuit implementation of complex-valued neural networks. Based on this, this paper proposes an in-memory computing circuit implementation of complex-valued Hopfield neural network (CHNN) for the first time, which provides a highly accurate and efficient processing circuit for portrait restoration. Firstly, a new memristive array is proposed, which can realize parallel complex-valued multiplication and complex-valued vector-matrix multiplication. On the basis, a CHNN circuit that can perform large-scale recursive computations is designed. Due to the characteristics of in-memory computation, the computation speed and robustness have been improved when realizing portrait restoration. Different portrait restoration scenarios can be realized based on the programmability of the memristive array. Pspice simulation results show that the recovery speed of CHNN can reach the level of 0.1 ms, and the accuracy can reach above 97.00%. Robustness analysis shows that the circuit can tolerate a certain degree of programming error and has strong anti-noise performance.

22 January 2024 in IEEE Transactions on Circuits and Systems I: Regular Papers
Verified
Journal Article
Closed access
HFHaotian Fu
Haotian Fu
YHYulong Huang
Yulong Huang
TCTingran Chen
Tingran Chen
CFChenyi Fu
Chenyi Fu
HRHongwei Ren
Hongwei Ren
BCBojun Cheng
Bojun Cheng
Compute-in-memory (CIM) based on emerging nonvolatile memory (eNVM) is an effective way to deploy neural networks to low-power edge devices for both storage and computation. NVMs such as ReRAM have been widely used in CIM. Meanwhile, MRAM has higher read and write cycles, lower device and cycle variation and a lower bit error rate, making it equally attractive for storage. However, the high read current and low on/off ratio result in large energy consumption in MRAM read limiting its large-scale application in CIM. The spiking neural network (SNN) represents the information as sparse spike sequences and facilitates hardware to achieve low-power computing by taking advantage of its spatial-temporal sparsity. To further increase the input sparsity of SNN and reduce the read energy consumption, this paper proposes ADC-free, dual-spike (DS) -CIM macro, a spiking MRAM CIM macro driven by asynchronous dual spikes. Compared to the conventional rate coding, our dual-spike coding method uses only 2 spikes to encode the information without losing accuracy. Moreover, the event-driven feature allows the macro to have sub-nW static power consumption. Our DS-CIM macro achieves comparable or higher accuracy while maintaining very low energy consumption. Specifically, it achieves accuracies of 96.99%, 82.87%, 90.00%, and 85.97% for digit classification, image classification, gesture recognition, and action recognition tasks, with energy consumption of only 8.07nJ, 71.26nJ, 729.3nJ, and 369.82nJ, respectively. These results emphasize the significance of DS-CIM and provide ideas for low-power inference on edge devices.

26 October 2023
Verified
Conference Paper
Open access
HRHongwei Ren
Hongwei Ren
YZYue Zhou
Yue Zhou
HFHaotian Fu
Haotian Fu
YHYulong Huang
Yulong Huang
RXRenjing Xu
Renjing Xu
BCBojun Cheng
Bojun Cheng
Event cameras have gained popularity in computer vision due to their data sparsity, high dynamic range, and low latency. As a bio-inspired sensor, event cameras generate sparse and asynchronous data, which is inherently incompatible with the traditional frame-based method. Alternatively, the point-based method can avoid additional modality transformation and naturally adapt to the sparsity of events. Still, it typically cannot reach a comparable accuracy as the frame-based method. We propose a lightweight and generalized point cloud network called TTPOINT which achieves competitive results even compared to the state-of-the-art (SOTA) frame-based method in action recognition tasks while only using 1.5 % of the computational resources. The model is adept at abstracting local and global geometry by hierarchy structure. By leveraging tensor-train compressed feature extractors, TTPOINT can be designed with minimal parameters and computational complexity. Additionally, we developed a straightforward downsampling algorithm to maintain the spatio-temporal feature. In the experiment, TTPOINT emerged as the SOTA method on three datasets while also attaining SOTA among point cloud methods on all five datasets. Moreover, by using the tensor-train decomposition method, the accuracy of the proposed TTPOINT is almost unaffected while compressing the parameter size by 55% in all five datasets.

17 April 2024 in IEEE Transactions on Circuits and Systems I: Regular Papers
Verified
Journal Article
Closed access
TCTingran Chen
Tingran Chen
WWWenjia Wang
Wenjia Wang
JCJiaqi Chen
Jiaqi Chen
HFHaotian Fu
Haotian Fu
WYWente Yi
Wente Yi
BPBiao Pan
Biao Pan
The large number of multiply accumulate (MAC) operations in Convolutional Neural Network (CNN) leads to substantial data migration and computation. Although computing-in-memory (CIM) proves to be a promising paradigm for MAC operations, high throughput CNN accelerator still confronts bottlenecks from: the low MAC utilization and the uncessary off-chip memory access. In this paper, we propose a high throughput CIM-based CNN accelerator PipeCIM with three hierarchies of pipelines: Intra-Macro, Near-Memory and Tile-Level. The Intra-Macro Pipeline parallelly executes data transfer and in-memory-computing (IMC) operations. The Near-Memory Pipeline alleviates memory access for pooling and data reshaping. The Tile-Level Pipeline establishes a layer-wise pipeline to further improve the throughput while reducing control complexity. PipeCIM introduces the nested scheme and a Unidirectional Divergent Connection Protocol (UDTCP) to simplify the control of data flow with the help of customized RISC-V instructions. To validate our design, PipeCIM was prototyped in 55 nm process node, achieving energy efficiency of 133.8 TOPS/W and peak throughput of 819 GOPS with a 16KB CIM array, which can accelerate VGG-16 to 128.56 $\times$ or Inception to 19.754 $\times$ compared to the baseline.