A team of researchers has designed and built a new chip, called NeuRRAM, which promises to turbo artificial intelligence : small and versatile, it consumes less power than conventional chips and performs accurate calculations directly in memory, without the need to rely on a external cloud and thus perform sophisticated cognitive tasks anywhere and anytime without relying on a network connection to a centralized server. This allows for a wider range of applications and offers a double benefit: better protection of sensitive data and the ability to run AI algorithms on a wider spectrum of edge devices .
A world of applications based on Artificial Intelligence
Applications abound in every corner of the world and in every aspect of our lives and range from smart watches, VR headsets, smart headsets, smart sensors in factories and rovers for space exploration. The study was published in the journal Nature with the title "A compute-in-memory chip based on resistive random-access memory" .
Currently, artificial intelligence algorithms require a significant expenditure of energy, mainly because they are computationally expensive . Most artificial intelligence applications on edge devices involve moving data from the devices to the cloud , where various artificial intelligence software processes and analyzes it. Then the results are moved back to the device. This is because most edge devices are battery powered and, as a result, have only a limited amount of power that can be devoted to data processing , thus being limited to capturing data and / or displaying results. "It's the equivalent of commuting for eight hours to a two-hour workday," said Wan, one of the researchers who worked on the project.
By reducing the power consumption required for AI inference at the edge, this NeuRRAM chip could lead to more robust, smarter and more accessible edge devices and smarter manufacturing . It could also lead to better data privacy as moving data from devices to the cloud carries greater security risks.
Another bottleneck in traditional computational architectures is the movement of data from memory to compute units. On AI chips, moving data from memory to compute units is a major bottleneck . To solve this data transfer problem, the researchers used what is known as resistive random access memory (RRAM), a type of non-volatile memory that allows computation directly within memory rather than in separate compute units. .
Computing with an RRAM chip is not necessarily new, but generally leads to a decrease in the accuracy of the calculations performed on the chip and a lack of flexibility in the chip architecture. The NeuRRAM chip, on the other hand, has been designed to be extremely versatile and supports many different models and architectures of neural networks. As a result, the chip can be used for many different applications, including image recognition and reconstruction, as well as speech recognition .
"Computing in memory has been a common practice in neuromorphic engineering since it was introduced more than 30 years ago (…) The novelty of NeuRRAM is that extreme efficiency now joins great flexibility for different artificial intelligence applications. with almost no loss of precision compared to standard digital computing platforms for general use ".
The article A new chip puts the turbo on the Artificial Intelligence for IoT devices was written on: Tech CuE | Close-up Engineering .