Researchers Make Hybrid Chips That Can Use AI To Run Smart Devices

JAKARTA - A group of researchers from Stanford developed a way to combine processors and memory in various hybrid chips that allow artificial intelligence (AI) to run battery-powered devices, such as smartphones and tablets.

The research team believes that various battery-powered electronics will be smarter if they are run using artificial intelligence algorithms. The obstacle is that efforts to build AI-capable chips for mobile are hampered by the “memory wall” as quoted from Slashgear, Tuesday, January 19

The memory wall is a term for the process of separating data and memory chips that must work together to meet the demands of AI computing.

Computer expert, Subasish Mitra, said that "Transactions between processors and memory can consume 95 percent of the energy needed to perform machine learning (ML) and AI, thus limiting battery life," said Subasish Mitra, author of the new study published in Nature Electronics .

The Stanford researcher designed a system that could run AI tasks more quickly and require less energy.

The researchers made use of a delam hybrid chip built in next to the self-storage memory. This new research builds on previous work developing a new memory technology called RRAM.

RRAM is capable of storing data with energy at a higher speed and energy efficiency than flash memory. The appearance of RRAM became the forerunner for further research, namely the development of hybrid chips.

The next important step was creating an algorithm that would allow eight separate hybrid chips to be combined into one energy-efficient AI processing engine.

The team claims they were able to "trick" the eight hybrid chips separately into feeling like a single chip. This system is known as the "Illusion System". They say that the eight chip system is only the beginning.

Simulations show that a system with 64 hybrid chips can run AI applications seven times faster than current processors using only one seventh of the energy.

The researchers believe that the performance of the prototype system shows that the Illusion System could be ready for market in the next 3 to 5 years.