When you purchase through contact on our site , we may earn an affiliate commission . Here ’s how it works .
Scientists inChinahave built a fresh type of tensor processing unit ( TPU ) — a special type of computer cow chip — usingcarbon nanotubesinstead of a traditional silicon semiconductor . They say the unexampled chip could give the door to more energy - efficientartificial intelligence(AI ) .
AI models are staggeringly data - intensive and require massive amount of computational power to consort . This acquaint a pregnant obstruction to grooming and scale up machine learning models , specially as the demand for AI diligence originate . This is why scientist are working on have young component — from processors tocomputing remembering — that are design to take in orders of magnitude less energy while melt the necessary reckoning .
Unlike conventional TPUs, this new chip is the first to use carbon nanotubes — tiny, cylindrical structures made of carbon atoms arranged in a hexagonal pattern — in place of traditional semiconductor materials like silicon.
Google scientists produce the TPUin 2015 to cover this challenge . These specialized chips act as consecrate hardware accelerators for tensor operations — complex mathematical calculations used to school and scat AI fashion model . By offloading these job from the fundamental processing unit ( CPU ) and artwork processing unit ( GPU ) , TPUs enable AI fashion model to be trained faster and more efficiently .
Unlike conventional TPUs , however , this new silicon chip is the first to use carbon nanotubes — tiny , cylindrical complex body part made of carbon particle coiffe in a hexagonal pattern — in place of traditional semiconductor materials like silicon . This structure allows electrons ( charged particles ) to flow through them with minimal resistance , piss carbon paper nanotube excellent music director of electricity . The scientists publish their research on July 22 in the journalNature Electronics .
tie in : Razor - sparse crystalline motion picture ' built atom - by - atom ' engender electrons moving 7 metre faster than in semiconductor unit
According to the scientist , their TPU consumes just 295 microwatts ( μW ) of king ( where 1 W is 1,000,000 μW ) and can deliver 1 trillion operations per watt — a unit of energy efficiency .
" From ChatGPT to Sora , artificial intelligence operation is show in a new revolution , but traditional Si - base semiconductor unit applied science is more and more unable to meet the processing needs of massive amount of data,“Zhiyong Zhang , conscientious objector - author of the newspaper and prof of electronics at Beijing ’s Peking University , toldTechXplore . " We have ground a resolution in the face of this global challenge . "
The new TPU is composed of 3,000 carbon carbon nanotube electronic transistor and is built with a systolic regalia computer architecture — a electronic web of central processing unit arrange in a grid - corresponding approach pattern .
— alone electronic transistor ' could commute the world of electronics ' thanks to nanosecond - musical scale flip speeds and refusal to wear out
— Intel unveils largest - ever AI ' neuromorphic computer ' that mimics the human genius
— ' Crazy thought ' memory gadget could slash AI energy consumption by up to 2,500 time
Systolic arrays pass on data through each C.P.U. in a synchronize , gradation - by - stride sequence , standardised to token moving along a conveyor knock . This enables the TPU to execute multiple calculations at the same time by coordinate the flow of data and assure that each central processor works on a small part of the task at the same prison term .
This parallel processing enable computing to be performed much more rapidly , which is crucial for AI role model processing large amounts of data . It also reduces how often the memory — specifically a case called static random - access memory ( SRAM ) — need to record and write data , Zhang said . By denigrate these operations , the new TPU can perform reckoning quicker while using much less muscularity .
To try out their new chip , the scientists build a five - level neural meshwork — a collection of machine watch algorithm designed to mimic the structure of the human brain — and used it for image identification task .
The TPU achieved an truth pace of 88 % while preserve power consumption of only 295 μW. In the future , like carbon nanotube - found technology could bring home the bacon a more energy - efficient choice to silicon - base chips , the researchers said .
The scientists plan to continue refining the chip to improve its functioning and make it more scalable , they said , including by explore how the TPU could be integrated into atomic number 14 CPUs .
