Hardware Implementation of Deep Network Accelerators Towards Healthcare and Biomedical Applications

Abstract

The advent of dedicated Deep Learning (DL) accelerators and neuromorphic processors has brought on new opportunities for applying both Deep and Spiking Neural Network (SNN) algorithms to healthcare and biomedical applications at the edge. This can facilitate the advancement of medical Internet of Things (IoT) systems and Point of Care (PoC) devices. In this paper, we provide a tutorial describing how various technologies including emerging memristive devices, Field Programmable Gate Arrays (FPGAs), and Complementary Metal Oxide Semiconductor (CMOS) can be used to develop efficient DL accelerators to solve a wide variety of diagnostic, pattern recognition, and signal processing problems in healthcare. Furthermore, we explore how spiking neuromorphic processors can complement their DL counterparts for processing biomedical signals. The tutorial is augmented with case studies of the vast literature on neural network and neuromorphic hardware as applied to the healthcare domain. We benchmark various hardware platforms by performing a sensor fusion signal processing task combining electromyography (EMG) signals with computer vision. Comparisons are made between dedicated neuromorphic processors and embedded AI accelerators in terms of inference latency and energy. Finally, we provide our analysis of the field and share a perspective on the advantages, disadvantages, challenges, and opportunities that various accelerators and neuromorphic processors introduce to healthcare and biomedical domains.

A Resource Constrained Neural Network for the Design of Embedded Human Posture Recognition Systems

“…In this paper, a custom HW design of a tiny Fully Convolutional Network (FCN) is presented to implement a HPR system, which better combines high recognition accuracy and low-energy and low-area requirements with respect to the existent literature, in order to extend the application range. Indeed, the FCN achieves state-of-the-art recognition accuracy both for laying and sitting postures, by exploiting only pressure sensors grouped in a small area close to the FCN, according to the edge-computing paradigm [28,29], without any particular distribution strategy. The FCN implements an end-to-end classification by exploiting a base-2 quantization scheme for weights and binarized activations [30,31] to meet the optimal trade-off between high recognition accuracy, the number of mapped physical resources and low power consumption [32,33].…”

An electronic neuromorphic system for real-time detection of high frequency oscillations (HFO) in intracranial EEG

“…Besides not being able to solve a broad range of pattern recognition tasks by design, the use of subthreshold analog circuits renders the design of the neural network more challenging in terms of robustness and classification accuracy. Nonetheless, successful examples of small-scale neuromorphic systems have been recently proposed to process bio-signals, such as Electrocardiogram (ECG) or Electromyography (EMG) signals, following this approach [15][16][17][18] . However, these systems were suboptimal, as they required external biosignal recording, frontend devices, and data conversion interfaces.…”

Adaptive Extreme Edge Computing for Wearable Devices

“…When dealing with SNN, the multi-sensory features or raw-data need to be encoded and fused in spike sequences in order to fit the input modality of the spike-based neural network. Furthermore, encoding the information in spikes can also further attenuate the risk of catastrophic forgetting issue in conventional neural networks (Azghadi et al, 2020 ). For decision level fusion, a voting mechanism is typically needed to output the final result after receiving the decisions from different sources of sensors which may be processed by different networks with different algorithms (Li et al, 2017 ).…”

“…The same structure was implemented on the embedded GPU and the comparison was performed in terms of accuracy, power consumption, and latency showing that the neuromorphic chips are able to achieve the same accuracy with significantly smaller energy-delay product, 30× and 600× more efficient for Loihi and ODIN/MorphIC, respectively (Ceolini et al, 2020 ). The comparison was further extended in Azghadi et al ( 2020 ), where the same task was applied to Field Programmable Gate Array (FPGA) and memristive implementations. Results show that neuromorphic hardware presents approximately two orders of magnitude improvement in the energy-delay product when compared to their FPGA counterparts, which highlights the prospective use of such architectures in edge computing.…”

“…To solve this problem, Application Specific Integrated Circuit (ASIC) accelerators try to reduce the complexity of the structure by making the system more application specific and using clock gating and specific hardware structure which matches best to the structure of the mapped neural network to reduce power consumption through less memory read and data access (Cavigelli and Benini, 2016 ; Chen et al, 2016 ; Lee et al, 2019 ; Song et al, 2019 ). For a complete survey on the state-of-the-art ASIC accelerators for biomedical signals refer to Azghadi et al ( 2020 ).…”