Artificial intelligence is useful in multiple valuable systems, such as monitoring a patient's cardiac activity, predicting a device's lifespan through its vibrations, and even incorporating complicated facial recognition tasks in video surveillance systems.
However, AI-based technologies typically require an excessive amount of power. In addition, in most cases, these machines have to be permanently connected to a cloud, raising IT security, data protection, and even energy usage.
Energy-Efficient Artificial Intelligence
According to SciTechDaily, CSEM engineers might have found a solution to maneuver the issues mentioned above. The group successfully developed an AI with a system-on-chip, and it only uses solar power to function, totally disregarding the cloud.
The artificial intelligence system that the engineers at CSEM developed runs on a small battery -- a tiny solar cell. The device can do typical AI operations using its built-in chip rather than relying on the cloud.
On top of that, the AI uses a fully modular system. As a result, it can be customized to any application wherever real-time image and signal processing is needed.
It is incredibly convenient when the AI has to deal with sensitive data.
SciTech Daily mentioned that the CSEM engineers responsible for the energy-efficient AI would be presenting their machine at the 2021 VLSI Circuits Symposium in Kyoto.
The event will occur within June.
How Does the CSEM System-on-Chip Work?
SciTech Daily stated that the CSEM system-on-chip primarily works on an entirely new signal processing architecture that decreases the required power.
It is composed of an ASIC chip that features a RISC-V processor, also made at CSEM, together with two tightly paired machine-learning accelerators.
So, for example, it will have one for face detection and another one for face classification.
According to the report, the first is a BDT or binary decision tree. It is an engine that could perform simple, mundane tasks but cannot serve more complex tasks such as recognition operations.
The second accelerator is a CNN, or convolutional neural network. This engine is responsible for doing complex tasks, like recognizing faces and detecting words. However, it also consumes the most energy.
Using a two-tier data processing approach, CSEM engineers drastically decreased the system's need for excessive power.
This mainly occurred because the machine runs the first accelerator regularly, and reserves the second accelerator for complicated tasks.
Enhancing the Accelerators
To push through with their research, the CSEM engineers maximized the accelerators' performances to make them more adaptable to various applications that require time-based signal and image processing.
Their innovation provides a gateway for developers to produce an entirely new generation of devices that use their energy-efficient artificial intelligence system on-chip.
These devices will have the capacity to provide services in places where it would be difficult to charge, change batteries, and find power sockets.
This article is owned by Tech Times
Written by Fran Sanders