Contradicting the trend of ever-shrinking technology, one company has built the biggest semiconductor in the world with incredible computing power.
We’ve come to expect that the technology underpinning our society will only get smaller as it improves, but a company called Cerebras Systems has gone in the completely opposite direction to build a semiconductor that dwarfs anything else on the market.
According to TechCrunch, the Cerebras Wafer Scale Engine is the world’s largest semiconductor – 57 times bigger than Nvidia’s largest GPU and equivalent in size to an iPad. It also includes 1.2trn transistors, giving it 1,000 times the performance of other chips, with the purpose of supercharging and speeding up AI applications.
While many semiconductor manufacturers have tried to produce smaller chips for AI applications, Cerebras has argued that building this enormous chip is actually more beneficial. This is because attempting to connect a large number of small chips together results in considerable latencies that slow down an AI’s ability to learn.
With 400,000 cores, this new chip eliminates latencies and speeds up the ability to process data, with the added ability of switching between processing and memory in a much shorter space of time.
Today is the perfect day to share our first tweet as we have announced the largest chip ever built to accelerate AI compute! Read all about the Cerebras Wafer Scale Engine (WSE) via @WIRED and @tsimonite https://t.co/ATaBRnxnGD pic.twitter.com/dX0xetyX6X
— Cerebras Systems (@CerebrasSystems) August 19, 2019
Cerebras has said building the chip has been a major challenge in itself, with issues such as thermal expansion as the connectors tethering the chip to a motherboard need to be able to expand at the exact same rate in order to prevent cracking. This was achieved by forging a new material in the lab that could handle the changes in temperature.
Another issue the company claims to have overcome is that it has found a way to prevent entire chips being tossed out because of impurities. While this common occurrence could be absorbed by a manufacturer with a small, inexpensive chip, it is a different matter with a chip as large and complex as the Wafer Scale Engine.
However, how the company will be able to prevent such impurities in mass production is not yet known.
Writing in a blog post, semiconductor pioneer and partner at Eclipse Ventures, Pierre Lamond, said: “The world is waiting for AI to fulfil its potential and that can only happen with a dedicated chip. A chip designed from the ground up for AI work.
“Every once in a while, a technology company comes along and defines a generation: Cerebras will define the AI generation.”