Efficient Computer raises $60M to keep AI devices running for months on end – News

Efficient Computer Co. says it’s going to make the dream of low-energy artificial intelligence computing a reality after raising $60 million in early-stage funding today.

The Series A round was led by Triatomic Capital and saw participation from Eclipse, Overlap Holdings, Union Square Ventures, RTX Ventures, Toyota Ventures, Overmatch Ventures and others, bringing its total amount raised to date to $76 million.

The startup has chosen a fitting name, for it has developed an entirely new kind of computer processor called the Electron E1 that’s based on a new “spatial dataflow” architecture. The chip’s purpose is to minimize energy consumption, power virtually any kind of computing device and keep them running for days, weeks, months or potentially even years at a time.

That should make it highly useful in today’s energy-hungry computing environments. Modern central processing units and graphics processing units are known to be inefficient when it comes to energy consumption because of the way they prioritize high performance, low latency and massive throughput over power conservation. As a result, they’re constrained by architectural overhead, including complex control logic, high-speed data movement and the need to maintain a precise execution state.

The Efficient Fabric architecture gets around this by eliminating all of the unnecessary data movement and architectural overheads that are intrinsic to today’s CPU and GPU architectures. By intelligently distributing workloads and connecting instructions to reflect the application dataflow, it can achieve dramatic gains in terms of performance per watt, without impacting performance or sacrificing programmability.

Efficient co-founder and Chief Executive Brandon Lucia believes the Electron E1 is a major breakthrough. He says energy is the primary constraint of all modern computing hardware, and it’s becoming extremely problematic as AI workloads move from the cloud and into the physical world.

There are big benefits to being able to run AI algorithms locally on devices, including enhanced data privacy, low-latency responses and offline functionality. But doing so is rarely practical. Existing processors struggle to support AI within their power, thermal, battery lifetime and form-factor limits.

While fixed-function accelerators do offer some improvement, Lucia said they lack the flexibility needed to keep pace with the rapidly evolving nature of AI workloads. “The industry has responded to rising energy costs by layering many fixed-function accelerators into a typical SoC,” he said. “This specialized hardware approach works to support a narrow slice of today’s workloads, but it breaks down as software, models and applications continue to change.”

To get around this, Efficient developed the Electron E1 around an entirely new idea. “The most durable path forward is a truly general-purpose architecture that can evolve with software over time, while providing market-leading energy efficiency for a range of critical intelligence use cases,” Lucia said.

Lucia said the Electron E1 has obvious applications in areas such as industrial automation and critical infrastructure, where energy efficiency often leads to expensive maintenance cycles and the need for constant manual oversight. For example, it will unlock the possibility of intelligent, battery-powered sensors that can run for years on a single charge. It can also support new applications in space and satellite systems, enabling them to carry more sensors and perform advanced functions without increasing their payload or cost.

To ensure the Electron E1 can run any kind of program, Efficient has developed the effcc Compiler, which works with all major programming languages, including C and TensorFlow. All developers have to do is drop in their code, and the compiler will optimize it to run on its Fabric architecture. This means it can power almost any kind of application with minimal code changes.

Triatomic Capital General Partner Peter Zhou said Efficient’s processors will help to bring intelligence to an array of devices and applications that were previously thought to be inaccessible. “As we continue to see AI embedded across the physical world, Efficient’s architecture is the missing link in AI’s last-mile distribution problem,” he said.

Going forward, Efficient said the funds from today’s round will enable it to accelerate its roadmap, hire more engineers and developers and bring the Electron E1 to new applications in emerging, AI-driven markets.

Image: Efficient Computer