Low-energy chip startup Efficient Computer Co. said today it has closed on a $97 million round of funding in the second major investment it has picked up this year.

The round was led by TQ Ventures and saw participation from Eclipse, Union Square Ventures, Giant Ventures, Triatomic Capital, TO Capital, TF Capital, Mana Ventures, ​Toyota Ventures, Overmatch and Borderless. It brings the company’s total amount raised so far to $650 million.

Efficient, which closed on a $60 million round in February, is in the process of bringing to market an entirely new kind of energy-efficient computer chip based on a data flow architecture. It claims that they can potentially reduce energy consumption by up to 100 times compared with the x86 chip architecture employed by companies such as Intel Corp. and Advanced Micro Devices Inc.

According to Efficient, its chips minimize energy consumption to such an extent that they could theoretically power devices for months or even years at a time. If Efficient really can do this, its chips are going to prove exceptionally useful in today’s energy-hungry artificial intelligence environments.

Modern central processing units and graphics processing units are extremely 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 overheads, including complex control logic, high-speed data movement and the need to maintain a precise execution state.

In a blog post announcing today’s round, Efficient co-founder and Chief Executive Brandon Lucia said the specialized processors used to accelerate AI workloads today are a “devil’s bargain” because of the way they sacrifice programmability and adaptability to increase their efficiency and speed. They’re only useful for one very narrow subset of tasks, namely AI inference, which means they’re useful for general-purpose computation.

Efficient’s data flow architecture eliminates 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.

Data flow chips have been conceptualized in academic literature for decades, but until now they have never seen commercial traction because of the difficulty of programming them for the wide variety of tasks that chips from Intel, AMD and Nvidia can handle. However, the desire for more energy-efficient AI processors has sparked renewed interest in the idea.

Lucia says Efficient took the basic concepts of the data flow architecture and then went back to the drawing board to design both the chips and the software tools needed to make them as adaptable as today’s general-purpose processors. Though it ultimately wants to make larger chips that can power data center servers, its initial target market is more modest.

Its first chips are designed to power small robots and autonomous drones that run on batteries as opposed to a power supply. “We sort of thread the needle where we’re easy to program, fast and efficient,” Lucia told Reuters in an interview. “When you build an AI system, the system ends up doing a lot more than two little nano-optimized AI algorithms.”

Efficient has not disclosed the names of its customers or details of its revenue, but Lucia insisted that the company is seeing “overwhelming demand” for its first product, the Electron E1 chip. The funds from today’s round will help Efficient to scale production volumes and increase shipments to its customers.

TQ Ventures partner Andrew Marks said he’s backing Efficient because the need for greater power-efficiency extends beyond just running AI models. “What convinced was Efficient’s ability to build both the hardware and the software, and turn that breakthrough into a business,” he explained. “Not only have they taped out four times, but they’re already shipping chips to customers at volume.”

Image: Efficient Computer