Here’s how Taiwan-based Upbeat Technology turned a 50-person sensor design company into a maker of a low-power microcontroller for edge AI applications in partnership with SiFive. The coming “secret weapon” on Upbeat’s roadmap is a compute-in-memory engine, now in prototype, planned for use in edge AI PCs.
CEO Jerry Chen, speaking in an interview with Fierce, said his company’s biggest challenge is making enough revenue to become a public company. Founded in 2021, the startup has received about $20 million in external funding. But Upbeat needs its own revenue, which Chen projects could reach $20 million in 2026. To get there, he expects to sell more of an already-popular vibration sensor in the form of a bone conduction microphone for high-quality sound capture accomplished by limiting environmental noise in headsets, smart glasses and earbuds.
“As a startup, we need to go IPO and in order to IPO we need revenue and profit, and the strategy is to use the sensor as the cash cow, and it looks good…It is feasible to say that in 2026, I can get $15 million to $20 million in revenue,” Chen said. The sensor sells for $1 apiece and fits in a market with few competitors. Smart glasses alone promise big growth for the device, with more than 50 companies in China making smart glasses.
The bone conduction microphone concept came out of MEMS sensor technology invented by a team of PhDs on staff at Upbeat. At its basic level, if functions as a vibration sensor for finding tiny vibrations in the ear canal. Chen said demand is high and Upbeat is racing to fill the purchase orders for the sensor, with the “best signal-to-noise ratio in the world.”
Upbeat's new MCU with SiFive RISC-V processors
In October, Upbeat and SiFive announced the UP201/UP301 microcontroller family designed for ultra-low power efficiency. It integrates two SiFive Essential RISC-V IP 32-bit embedded processors and two AI accelerators by Upbeat. The intended audiences are always-on IoT devices such as wearables, drones and AI inference applications, even voice-interactive toys. It includes Upbeat’s patented Error Detection and Correction architecture.
Drones could be a sweet spot for Upbeat, since the microcontroller will provide inertial guidance should a drone’s connection to GPS be blocked. Drones need to sense multiple variables that the microcontroller supports, including temperature, pressure, velocity, distance, and vibration. The microcontroller works with any sensor maker’s products, Chen said.
Because the majority of drones globally are made in China and are subject to potential tariffs imposed by the US, Upbeat can take advantage of its home location in Taipei, Taiwan, where Chen said the US and Taiwan are working to develop Taiwan-based drone alternatives to the Chinese supply chain. The UP201/UP301 microcontroller would in effect become a sensor hub for drones, he added.
Aside from the bone conduction microphone sensor and the microcontroller, Upbeat is in prototype stage for a compute-in-memory chip that will be ideal for tiny, low-power chips embedded in fingers in robots where dexterity is essential. “Robot fingers cannot use a power-hungry chip,” Chen said. “It’s our secret weapon.”