2024 BrainChip Discussion, page-9758

BringingTouch to the Edge: A Neuromorphic

ProcessingApproach For Event-Based Tactile

Systems

Harshil Patel

Brainchip Research Institute

Perth, Australia

Email: [email protected]

Anup Vanarse

Brainchip Research Institute

Perth, Australia

Email: [email protected]

Kristofor D. Carlson

Brainchip Inc.

Laguna Hills, USA

Email: [email protected]

Adam Osseiran

Brainchip Research Institute

University of Western Australia

Perth, Australia

Email: [email protected]

Abstract— The rise of neuromorphic applications has

highlighted the remarkable potential of biologically-inspired

systems. Despite significant advancements in audio and visual

technologies, research directed towards tactile sensing has not

been as extensive. We propose a neuromorphic tactile system

for sensing and processing that presents promising results for

edge devices and applications. In this study, a neuromorphic

tactile sensor, two data encoding techniques, and a two-layer

spiking neural network (SNN) deployed on the AKD1000

Akida Neuromorphic System on Chip (NSoC) were used to

demonstrate the system's capabilities. Results from

experiments on the ST-MNIST dataset showed high accuracy,

with the complement-coded variant achieving 93.1%,

outperforming previous state-of-the-art models for this dataset.

Additionally, an exploratory study showed that early

classification was possible, with most samples requiring only

38% of the available events to classify correctly, reducing the

amount of data that needs to be processed. The low power

consumption and high throughput of both SNN models, with

an average dynamic power consumption of 6.37 mW and 7.76

mW and an average throughput of 586 and 589 frames-per-

second respectively, make the proposed system suitable for

edge devices with limited power and processing resources.

Overall, the proposed tactile sensing system presents a

promising solution for edge applications that require high

accuracy, low power consumption, and high throughput….

This study suggests that the proposed early-classification

pipeline holds the potential to significantly reduce

classification latency for real-time systems, thus avoidingthe

trade-off between sample capture time and accuracy in time-

critical environments. For instance, in applications such as

autonomous vehicle control, faster inferences could be

critical, and making inferences during data capture could

provide intermediary results before a final classification is

made with the complete data sample. By reducing the latency

associated with the data collection window, this system could

enable real-world systems to make classifications with even

incomplete data before making a final high-confidence

classification.

V.CONCLUSION

The proposed neuromorphic system for tactile sensing

and processing presents promising results, offering a solution

for edge devices. In this study, two data encoding techniques

were used in combination with a two-layer SNN deployed on

the Akida NSoC. Results from experiments on the ST-

MNIST dataset showed high accuracy, with the complement

coded variant achieving 93.1%, outperforming previous

state-of-the-art models for this dataset. Additionally, it was

found that early classification was possible, as most samples

could be correctly classified with just 38% of the available

events, thus enabling real-time systems to reduce latency

induced by data collection. The low power consumption and

high throughput of both SNN models, with average dynamic

power consumption of 6.37 mW and 7.76 mW, and average

throughput of 586 and 589 frames-per-second respectively,

make this system suitable for edge devices with limited

power and processing resources. In conclusion, the proposed

tactile sensing system presents a promising solution for edge

applications, with high accuracy, low power consumption,

and high throughput.”

https://ieeexplore.ieee.org/abstract/document/10168592/

I have highlighted in red a most significant feature of AKIDA technology.
This characteristic is very human. We all when driving see things ahead
of us and without waiting to be completely sure that it is a person in an Italian
Suit complimented by a Gucci tie we make the decision that it is human
and could present a hazard and we react by reducing speed, hovering over
the brake pedal, changing lanes etc.

In other words we do not require 100% of the data to start activating our response
mechanism. For us as human drivers even a shadow of a human starting to appear
can be enough data.

AKIDA unlike Von Neuman compute which needs to have a whole image of the
object to process, it can just like a human commence to process the data coming
from the camera and reach correct classifications with only 38% of the incoming
data.

We all know that driving safely is about having sufficient time to react to whatever
occurs because no matter what a ton of metal does not stop instantly it always
takes some time no matter who or what is at the wheel and the more time who or
what has the better the outcomes.

The revolutionary nature of AKIDA technology continues to confound even before
the original purpose of this paper comes into play which was as it does to prove
out a method for AKIDA technology to bring touch to robotics in the same way that
human skin does.

My opinion only DYOR

Fact Finder