2022 BRN Discussion, page-3011

On a serious note the following peer reviewed paper was published on 5 January, 2022 and for all those out there worried about competition in the form of Analogue SNN chips it basically states that they cannot scale are hard to train and have variable accuracy.

(These FACTS were known to Peter van der Made and Anil Mankar a decade ago and that is why a Digital implementation of SNN was pursued.)

The authors of this paper claim to have come up with a way to improve accuracy of analogue SNN. They now need to convince backers develop a prototype that works then design their chip have it engineered etc, etc so hardly anywhere near to being a commercial threat to Brainchip for years.

It goes without saying that they prove conclusively anyone out there flogging current analogue SNN are no competition for this the completely scalable, one shot training highly accurate AKIDA digital technology chips and IP.

My opinion only DYOR
FF

AKIDA BALLISTA



Improving Spiking Neural Network Accuracy Using Time-based NeuronsHanseok KimDept of ECE, ISRC Seoul National University Seoul, South Korea [email protected]—Due to the fundamental limit to reducing power consumption of running deep learning models on von-Neumann architecture, research on neuromorphic computing systems based on low-power spiking neural networks using analog neurons is in the spotlight. In order to integrate a large number of neurons, neurons need to be designed to occupy a small area, but as technology scales down, analog neurons are difficult to scale, and they suffer from reduced voltage headroom/dynamic range and circuit nonlinearities. In light of this, this paper first models the nonlinear behavior of existing current-mirror-based voltage- domain neurons designed in a 28nm process, and show SNN inference accuracy can be severely degraded by the effect of neuron’s nonlinearity. Then, to mitigate this problem, we propose a novel neuron, which processes incoming spikes in the time domain and greatly improves the linearity, thereby improving the inference accuracy compared to the existing voltage-domain neuron. Tested on the MNIST dataset, the inference error rate of the proposed neuron differs by less than 0.1% from that of the ideal neuron.Index Terms—Artificial neural network, spiking neural net- work, time-based signal processing, integrate-and-fire neuron, ANN-to-SNN conversion