Selected Projects

	Abstract—Cloud-based training and edge-based inference modes for Artificial Intelligence of Medical Things (AIoMT) applications suffer from accuracy degradation due to physiological signal variations among patients. On-chip learning can overcome this issue by online adaptation of neural network parameters for user-specific tasks. However, existing on-chip learning processors have limitations in terms of versatility, resource utilization, and energy efficiency. We propose HybMED, which is a novel neural signal processor that supports on-chip hybrid neural network training using a composite direct feedback alignment-based paradigm. HybMED is suitable for general-purpose health monitoring AIoMT devices. It improves resource utilization and area efficiency by the reconfigurable.
	Versatile and energy-efficient neural signal processors are in high demand in brain-machine interfaces and closed loop neuromodulation applications. In this paper, we propose an energy-efficient processor for neural signal analyses. The proposed processor utilizes three key techniques to efficiently improve versatility and energy efficiency. 1) Hybrid neural network design: The processor supports artificial neural network (ANN)- and spiking neural network (SNN)-based neuromorphic processing where ANN is used to support the processing of ExG signals and SNN is used for handling neural spike signals. 2) Event-driven processing: The processor can perform always-on binary neural network (BNN)-based event detection with low-energy consumption, and it only switches to the high-accuracy convolutional neural network (CNN)-based recognition mode when events are detected. 3) Reconfigurable architecture: By exploiting the computational similarity of different neural networks, the processor supports critical BNN, CNN, and SNN operations with the same processing elements, achieving significant area reduction and energy efficiency improvement over those of a naive implementation. It achieves 90.05% accuracy and 4.38 uJ/class in a center-out reaching task with an SNN and 99.4% sensitivity, 98.6% specificity, and 1.93 uJ/class in an EEG-based seizure prediction task with dual neural network-based event-driven processing.
	Abstract—Low power consumption associated with data transmission and processing of wearable/implantable devices is crucial to ensure the usability of continuous health monitoring systems. In this paper, we propose a novel health monitoring framework where the signal acquired is compressed in a task-aware manner to preserve task-relevant information at the sensor end with a low computation cost. The resulting compressed signals can be transmitted with significantly lower bandwidth, analyzed directly without a dedicated reconstruction process, or reconstructed with high fidelity. Also, we propose a dedicated hardware architecture with sparse Booth encoding multiplication and the 1-D convolution pipeline for the task-aware compression and the analysis modules, respectively. Extensive experiments show that the proposed framework is accurate, with a seizure prediction accuracy of 89.70% under a signal compression ratio of 1/16. The hardware architecture is implemented on an Alveo U250 FPGA board, achieving a power of 0.207 W at a clock frequency of 100 MHz.