Traditional processors suck down a lot of juice, especially at scale. They also do it in a horribly inefficient way, according to Carnegie Mellon professor of electrical and computer engineering Brandon Lucia.
Lucia is the CEO and co-founder of Efficient Computer, a startup claiming its unique Fabric architecture is ten to 100 times as power efficient as processors using the reigning von Neumann architecture, which dates to 1945 a!nd relies on sequential processing. At the company’s booth at CES 2025, Lucia told IT Brew the Fabric architecture derives from an “ideological goal.”
Lucia said the Efficient Computer team approached the design of Fabric with a central principle: “Every time we’re at a decision point in this design, we’re going to opt for the most energy-efficient option… What we ended up with is something that doesn’t really look like a processor in the way that processors have been for a long time.”
Traditional processors do sequential operations with branch prediction, and between each operation the processor needs to refer to memory adjacent to the processing pipeline and perform configuration changes, Lucia said: “On every single cycle, you’re doing that multiple billions of times per second, that’s incredibly wasteful.”
Fabric instead spatially arrays them in a grid of processing elements called a data flow graph, allowing more parallel processing and cutting out “a lot of the needless overheads” of von Neumann processors, according to Lucia.
“Each operation corresponds to one of those squares in the grid, and there are connections between the operations in your program,” Lucia said, adding Fabric’s compiler and software stack can extract that data from any program using a general-purpose language.
“We configure the hardware so there’s a direct hardware path between operations that interact with one another, and we have an extremely energy efficient network on-chip that allows data to flow from a producer operation to a consumer operation,” he said. “The data flow abstraction is fully general.”
Top insights for IT pros
From cybersecurity and big data to cloud computing, IT Brew covers the latest trends shaping business tech in our 4x weekly newsletter, virtual events with industry experts, and digital guides.
Lucia isn’t dinging the von Neumann design, he said—numerous advances have made them far more efficient, and the overhead resulted from the compromises necessary to design a processor theoretically capable of solving any mathematical task (a remarkable breakthrough in computing). However, Lucia argued Fabric doesn’t need to make a “major concession” on performance due to increased parallelism.
Efficient Computer is focusing on the embedded CPU market, where Lucia said their designs achieve performance comparable to or better than the competition. For example, the firm is currently developing proof of concept sensors to demonstrate Fabric’s potential in industrial automation.
Sectors he’s particularly excited about include space and defense, where Lucia said there’s a need for processors that consume dramatically less power. Think CubeSats, or freeing future infantry from the need to carry weighty battery packs to power future tech like heads-up displays. The company has also been quick to point out Fabric might be very useful in edge AI.
“There’s a lot more computing happening on orbit today,” Lucia said. “If you look back 10 years, that was very exotic. Now it’s becoming commonplace.”
In the future, Lucia told IT Brew he hopes the Fabric architecture will prove scalable and competitive for applications like mobile, edge cloud, or even the data center.
“The architecture is not tied fundamentally to small devices,” he said. “That’s just where we saw a huge opportunity, obvious, immediate, near-term impact.”