Technologies

Discover, Connect & Collaborate at TECHINNOVATION 2021

An End-to-end Approach to Enable Deep Learning at the Edge for Ultra-low Power AI Systems

Technology Overview

As computing devices nowadays become ever more multi-functional with a wide plethora of features, the need for an energy-efficient computing system has become increasingly crucial. We offer an approach for deep learning at the edge for ultra-low power AI systems, where the entire development stack, from algorithms to hardware devices, was optimised. The main goal of this approach is to enable the execution of deep learning applications on ultra-low power edge platforms, while conserving the battery lifetime. This approach achieves significant gains over current solutions, reaching 10X savings on total system power.

Technology Features, Specifications and Advantages

We offer a full-system solutions that enables artificial intelligence on ultra-low-power edge platforms. Our approach optimizes the entire computing stack for edge platforms: from algorithms to nanodevices. A key

differentiator is our use of low-power non-volatile memory technologies such as resistive RAM (RRAM) and magnetic RAM (MRAM). These memories technologies have recently begun to be commercially offered by major semiconductor foundries. Having performed extensive research on these technologies in the past, with knowledge of the inner workings of each type of memory, we are poised to have a significant competitive edge against other companies with similar approaches. Another key differentiator is our full stack optimizations, where we perform software-based transformations and design specific hardware modules to enhance this transformation for better efficiency. We project to enable AI applications in a much smaller form-factor than what is possible today while extending the battery life by 10X in low-power (<1Watt) IoT platforms, compared to our competitors.

Potential Applications

Our main target customer is in the IoT sensor domain, health monitoring and wearables in general. Other areas include the following:

  • Hearables
  • Smart buildings
  • Infrastructure sensor/meters
  • Smart agriculture
  • Computer vision (tiny applications)

Our disruptive technology enables running AI workloads at the edge while retaining battery life. The projected market size for AI edge hardware is expected to reach $6.5 Billion by 2025, and the edge computing market size is expected to reach $16 Billion by 2025.

Customer Benefit

This platform allows customer to map deep learning and AI on very low power platforms. This would significantly aid in better event detection and prediction, without the need for communication with the cloud. As such, it enables a new capability at ultra-low power consumption, which was not possible before.

This edge AI full system solution would benefit many areas and industries and even infrastructure projects. This platform would not only develop products to meet user demands but could also be a catalyst in developing talents in the field of edge AI systems.

OVERVIEW
Contact Person

Jianxin Yao

Organisation

NTU - NTUitive >> ICT, AgriTech and FoodTech

Technology Category

  • Infocomm
  • Artificial Intelligence, Internet of Things & Wearable Technology

Technology Readiness Level

Keywords

Deep learning at the edge for ultra-low power AI systems