AHEAD: Automatic Holistic Energy-Aware Design Methodology for MLP Neural Network Hardware Generation in Proactive BMI Edge Devices

Abstract

The prediction of a high-level cognitive function based on a proactive brain–machine interface (BMI) control edge device is an emerging technology for improving the quality of life for disabled people. However, maintaining the stability of multiunit neural recordings is made difficult by the nonstationary nature of neurons and can affect the overall performance of proactive BMI control. Thus, it requires regular recalibration to retrain a neural network decoder for proactive control. However, retraining may lead to changes in the network parameters, such as the network topology. In terms of the hardware implementation of the neural decoder for real-time and low-power processing, it takes time to modify or redesign the hardware accelerator. Consequently, handling the engineering change of the low-power hardware design requires substantial human resources and time. To address this design challenge, this work proposes AHEAD: an automatic holistic energy-aware design methodology for multilayer perceptron (MLP) neural network hardware generation in proactive BMI edge devices. By taking a holistic analysis of the proactive BMI design flow, the approach makes judicious use of the intelligent bit-width identification (BWID) and configurable hardware generation, which autonomously integrate to generate the low-power hardware decoder. The proposed AHEAD methodology begins with the trained MLP parameters and golden datasets and produces an efficient hardware design in terms of performance, power, and area (PPA) with the least loss of accuracy. The results show that the proposed methodology is up to a 4X faster in performance, 3X lower in terms of power consumption, and achieves a 5X reduction in area resources, with exact accuracy, compared to floating-point and half-floating-point design on a field-programmable gate array (FPGA), which makes it a promising design methodology for proactive BMI edge devices.

Echo State Networks for Estimating Exteroceptive Conditions From Proprioceptive States in Quadruped Robots

“…Problems related to efficiency in terms of power, performance, and occupied area were also considered and addressed. This approach is in line with another recent example that follows this hardware-oriented strategy (Huang et al, 2020): here an automatic holistic energy-aware design methodology is proposed and applied to a multilayer perceptron designed to be embedded in proactive brain-machine interface edge devices based on FPGA. Another interesting direction for hardware implementation is related to the opensource Neural Network framework called Neural Network on Microcontroller (NNoM) 1 , for implementing (recurrent) neural networks on a microcontroller.…”

A brain machine interface framework for exploring proactive control of smart environments

“…To evaluate further optimization potential for the energy consumption of the presented type of decoder, we compared our hardware implementation of an adaptable decoder with an optimized implementation of a non-adaptable decoder for an FPGA. We found that the non-adaptable decoder exhibits an energy uptake which is reduced by a factor 2-3 only 25 , such that further optimisation of the FPGA implementation is not expected to yield significant gains.…”

Industrial Control under Non-Ideal Measurements: Data-Based Signal Processing as an Alternative to Controller Retuning

“…As a summary, our data-based enhanced control strategy, which is implemented by means of ANNs, allows us to improve a control approach by means of (i) decreasing the design process complexity, and (ii) increasing its scalability [ 14 , 16 ]. The complexity reduction is achieved due to the fact that ANNs do not require a such precise adjustment to the scenario as usual filtering strategies do.…”

“…Data-driven methods and ANNs have arisen as new approaches able to offer a good control performance at the same time they increase the scalability and decoupling of the control strategy from the highly complex mathematical models [ 14 , 15 , 16 , 17 , 18 , 19 ]. In such a context, ANNs have been considered to perform different tasks: (i) act as soft-sensors, (ii) complement the model-based controllers, and (iii) act as a control strategy as such.…”

“…For that reason, the approach proposed in [ 36 ] has been enhanced in this work by means of a complete and easy tuning data-based enhanced control solution. The new proposed approach will achieve the reduction of the design process complexity and the increment of its scalability [ 14 , 16 ]. It considers a new denoising approach, the data-based Denoising Stage.…”