Efficient On-Chip Learning of Multi-Layer Perceptron Based on Neuron Multiplexing Method

Abstract

An efficient on-chip learning method based on neuron multiplexing is proposed in this paper to address the limitations of traditional on-chip learning methods, including low resource utilization and non-tunable parallelism. The proposed method utilizes a configurable neuron calculation unit (NCU) to calculate neural networks in different degrees of parallelism through multiplexing NCUs at different levels, and resource utilization can be increased by reducing the number of NCUs since the resource consumption is predominantly determined by the number of NCUs and the data bit-width, which are decoupled from the specific topology. To better support the proposed method and minimize RAM block usage, a weight segmentation and recombination method is introduced, accompanied by a detailed explanation of the access order. Moreover, a performance model is developed to facilitate parameter selection process. Experimental results conducted on an FPGA development board demonstrate that the proposed method has lower resource consumption, higher resource utilization, and greater generality compared to other methods.

Achieving High Efficiency: Resource sharing techniques in artificial neural networks for resource-constrained devices

“…By means of multiplexing the inputs and the outputs this approach allows ANN to process multiple data streams in a nearly simultaneous manner. Another recent research [10] demonstrates the architecture of a configurable neuron calculation unit (NCU) that employs some level of processing nodes reduction through multiplexing NCUs. Binding this with limiting the data bit-width they try to increase the efficiency of the computation resource usage and lower the energy consumption.…”