- Common computing announces CN101, which is the world’s first thermodynamic computing chip
- Startup says its approach supports scaling of AI burden in existing data center power limits
- The purpose of future designs is to supply high performance within the current infrastructure
Common computing has announced a successful CN101’s successful tapeout, describing it as the “world’s first thermoditic computing chip”.
Startup views this development as a natural response to the growing energy requirements of AI and scientific work.
Unlike the CPU and GPUs that rely on precise logic, CN101 is designed to exploit natural dynamics such as volatility, consumption and random pin.
You can like
General Promise Roadmap
The idea is to accelerate certain reasoning tasks, while reducing the use of energy, drawing the processes that usually suppress the current chips.
The company says CN101 has been targeted in two specific types of calculation. One contains a large -scale linear algebra, which is central to correction issues and scientific modeling.
The second is a stockstick sampling, where the purpose of the normal random walking walking is to accelerate the data methods, including the Bisian Inconity.
“In recent months, we have seen that AI’s capabilities are approaching a flattened curved letter with today’s energy budget and architecture, even when we plan to measure training, 10,000 X is running in the next 5 years. The emerging example – an emerging example – a primarily executed by a small engineering team, said KCO Frees Sabahi.
Looking forward, the company has compiled a roadmap that starts with CN101 but extends over the next decade.
“Our vision for a scale of spread with our stockstick hardware begins with our vision for the CN101 demonstrating key applications on CN101, then getting the latest performance on medium -sized genius with CN201 next year, and eventually gaining CN301 with a massive CN301. The general computing scientist Patrick Coles explained.
Ordinary engineers say that the tapeout also represents the first step towards this feature on how these theories behave in the real silicon.
“CN101 represents the first silicon demonstration of our thermoddynten architecture that takes advantage of random, metastabilli, and noise to perform sample tasks. From the feature of CN101, we will be able to understand how this random process will support the real Silicon, and to support our plains, to support the state. Said
Common computing was founded in 2022 by engineers from Google Brain, Google X, and Pellantier.
