Startup raises $16M for power-saving chips
As Nvidia tops charts with a record $2 trillion evaluation for making the computer chips that power AI, a Pittsburgh startup is hoping to carve out its own piece of the chip pie by focusing on power.
Specifically, reducing the power that chips need to operate, which could extend the functional life of everything from smartwatches to satellites, and open up new opportunities for infrastructure monitoring and defense.
“We saw the promise in the product research,” said Brandon Lucia, co-founder and CEO of the East Liberty-based startup Efficient Computer, which closed an initial $16 million funding round last week led by Eclipse, a Silicon Valley venture group.
The company says its chips are 100 times more energy efficient than what is currently in the market, allowing systems to last longer on a single charge. According to its website, Efficient Computer’s processors can be powered by a single AA battery for up to a decade.
Mr. Lucia said the first market the company is targeting is technology that doesn’t need as much power: “smart sensors and wearables and devices that you might deploy into an environment for five to ten years.”
The simpler systems still need to be able to run a variety of computations that can include AI, machine learning, signal processing and analytics, he said.
Inefficient processors have previously slowed innovation and limited applications, Eclipse partner Greg Reichow said in a statement.
“More than just closing this gap,” he said, “the Efficient team is introducing an entirely new category of processor that is enabling organizations to reconsider what is possible.”
The startup grew out of seven years of research at Carnegie Mellon University, where Mr. Lucia and co-founder Nathan Beckmann
serve on the faculty. It now employs a team of about 20.
One big way that they save power is by limiting correspondence between the chip and cloud-based storage systems. Instead of sending all raw data to be processed offsite, Efficient’s chip does that work on location and only transmits the compressed finished product.
Researchers at Northwestern University found a similar hack last fall, gaining the same 100-times efficiency jump by cutting out the cloud. But that research, published in October in the journal Nature Electronics, relied on a type of manufacturing not yet offered by mainstream chipmakers, Mr. Lucia said.
“Efficient’s architecture can be implemented in any fabrication technology, including the ones on offer from the major chip fabs,” he said.
Efficient Computer’s growth comes during a broader push for more efficient chips, especially to power AI. A report over the weekend found that ChatGPT can draw as much power daily as 17,000 households. Tech giant IBM created its own chip last fall to cut down on power use.
Efficient is building for a more “general-purpose operation” but said its system can still be used to support AI software. Its underlying architecture “minimizes the cost of controlling the hardware” and moving data inside of the chip, Mr. Lucia said.