Mueon raises $65M Series A to build 3D server modules

Maverick Silicon, a semiconductor-focused fund, led the round. Intel's venture capital arm, Oregon Venture Fund, Wellington Capital, IQT and several other investors also participated.

Mueon was founded last year by Wilfred Gomes and Jack Hwang. Gomes, the company's chief executive officer, worked on die stacking technologies during his time at Intel. Hwang, Mueon's chief technology officer, helped develop the techniques Intel uses to etch transistors into silicon wafers.

A traditional server's core components sit next to one another on a motherboard in a two-dimensional configuration. Mueon's Cubelet architecture instead stacks computing and memory components vertically. According to the company, the design could prove up to 10 times more power- and cost-efficient than current hardware, because packing components closer together shortens the physical distance data must travel between them, which lowers associated power use and speeds up computations.

In a blog post published last year, Mueon investor Oregon Venture Fund described Cubelets as "Legos for the data center." The post stated that a Cubelet module includes not only computing and storage modules but also the auxiliary hardware needed to operate them reliably, including interconnect, power delivery and thermal management equipment.

Thermal management is a major challenge in 3D computing systems. Processors and memory chips often have different maximum operating temperatures, which means a processor running near its maximum can overheat memory modules stacked on top of it. In many servers, chips dissipate part of the heat they generate by channeling it to the motherboard, a relatively simple path when a chip sits directly on that board. In a 3D system, heat from the top layer must pass through multiple other components before reaching the motherboard.

Oregon Venture Fund's blog post said Mueon had found a way to address the thermal management issues of 3D systems. The post also detailed that the company's Cubelet modules can host multiple types of chips, including graphics cards and Arm central processing units.

Mueon said its Cubelets are optimized for artificial intelligence and hyperscale workloads. The company will use the proceeds from the funding round to hire more engineers and improve its system design. In parallel, it is developing software that will let customers run popular AI frameworks on its hardware.

"This financing gives us the scale to bring the next generation of compute and memory system-scale integration to market," Gomes said. "Mueon's architecture is built to move past today's rack-level constraints and deliver the kind of density the rest of the industry is only beginning to chart a path toward."