Overview

Reduced Energy Microsystems (REM) was a San Francisco-based fabless semiconductor startup founded in 2014 by William Koven, Eleazar Vega-Gonzalez, and Dylan Hand. ^[1] The company built ultra-low-power chips for embedded machine learning inference at the edge — targeting applications like police body cameras, security systems, drones, and augmented reality devices — using asynchronous circuit design, a decades-old academic approach that eliminates the master clock governing conventional chips. ^[2] REM entered Y Combinator's Summer 2015 batch as the accelerator's first chip startup. ^[3]

REM failed because asynchronous chip development — despite genuine technical merit — requires capital-intensive, multi-year development cycles that a $2M seed-funded startup could not sustain long enough to reach commercial scale. The edge AI inference market it was targeting simultaneously attracted well-funded incumbents and specialists who commoditized the space before REM could ship at volume.

Crunchbase lists REM as permanently closed as of August 21, 2020 — roughly six years after founding and approximately three years after its last documented public activity. ^[4] No acquisition, acqui-hire, or public post-mortem was announced. Koven subsequently joined Galois, Inc. as a research engineer before becoming a chip lead at Meta; Vega-Gonzalez went on to found Coil, a builder AI startup that entered YC's Winter 2023 batch. ^{[5] [6]}

Founding Story

Reduced Energy Microsystems emerged from a specific convergence of academic research, industry frustration, and a window in the semiconductor market that the founders believed incumbents were structurally unable to exploit.

William Koven, the CEO, came to the founding with direct industry experience. He held a B.S. in Engineering from Harvey Mudd College, where he was a Clay-Wolkin Fellow, and had spent years at Intel before co-founding REM. ^{[7] [8]} His time at Intel gave him a close-up view of why large semiconductor companies resist architectural change: the organizational and toolchain investments in synchronous design are so deep that alternatives — even technically superior ones — get rejected on institutional grounds. "Asynchronous is kind of a dirty word at Intel," Koven later said. "It's just too different from what they do as an organization. They've thrown out a lot of technologies like this because they're different and scary." ^[9]

Dylan Hand was a Ph.D. student at USC's Viterbi School of Engineering, where he worked in the E2S2C (Energy-Efficient Systems, Circuits, and Chips) lab — the academic home of asynchronous VLSI research at USC. ^[10] The lab had a prior commercialization lineage: TimeLess Design Automation, an asynchronous EDA startup spun out of the same research group, had been acquired by Fulcrum Microsystems, which was itself acquired by Intel in 2011. ^[11] That chain of acquisitions suggested that asynchronous chip IP had real value — but also that the path to standalone company success in this space was historically difficult.