Aspinity

Aspinity designs analog processing chips for always-on sensing and signal classification. Its AnalogML architecture, built on the RAMP (Reconfigurable Analog Module Processor) platform, performs detection and inference directly on raw analog sensor data, keeping the MCU and other digital processors asleep until meaningful data are detected. The company describes this as an "analyze-first" approach intended for the large population of devices that continuously monitor real-time sensor streams, where conventional digital processing spends power digitizing and evaluating data that is mostly irrelevant.

The product line consists of the AML100 analog machine learning core, which the company states is shipping and being designed in, and the AML200, described as in development and scaling performance and applications. Aspinity positions the same analog core across several application areas: counter-drone/cUAS detection (marketed as SkyML) using battery-powered nodes said to provide more than three minutes of early warning for RF-silent drones; wideband RF sensing and signal classification performed at the antenna, before the ADC; industrial IoT and asset-protection anomaly detection running on-device without cloud connectivity or custom training, cited at 3+ years of always-on operation on a battery; and security applications spanning smart home, commercial property, fleet and vehicle, cited at under 100 µA of system draw.

The company's team is described as combining analog and mixed-signal semiconductor leadership with the academic researchers who developed the AnalogML architecture, with prior roles at Cadence, Micron, Qualcomm, Vesper, Bosch/Akustica, STMicroelectronics and Draper Laboratory. Aspinity cites decades of R&D and a broad patent portfolio addressing variability and repeatability challenges in analog circuitry.

Founding story

Aspinity's technology originates in Ph.D. research at Georgia Tech and West Virginia University. David Graham, a WVU associate professor of computer science and electrical engineering, received an NSF CAREER award for research on energy-efficient sensor networks using analog signal processing; he had earlier modeled the cochlea with neuromorphic silicon circuits at Georgia Tech under advisor Jennifer Hasler, herself a student of neuromorphic computing pioneer Carver Mead. WVU alumnus and business executive Tom Doyle contacted Graham about commercializing the work, and the founders came together in 2012. With Graham's former doctoral student Brandon Rumberg, who became co-founder and CTO, and early funding from WVU, the team prototyped analog chips for always-listening voice and acoustic event detection. A licensing agreement with WVU in 2015 is credited by Doyle with starting the company in earnest; the company's 2023 press release also states Aspinity was founded in 2015.

Business model

Aspinity develops and sells analog machine learning semiconductor products (the AML100, with the AML200 in development) to device makers, deriving revenue from silicon that is designed into customer products. Its industrial offering is described as requiring no recurring cost. Distribution and market access are pursued partly through strategic partners and distributors, such as automotive semiconductor distributor Unitrontech.

Sales of analog machine learning chips designed into customer products; the industrial asset-protection offering is described as having no recurring cost.

Traction

The AML100 is described as shipping and being designed in, with the AML200 scaling performance and applications. The company raised more than $19 million in total funding as of September 2023 and has partnerships with Infineon Technologies, STMicroelectronics and Unitrontech.

Latest developments

The company's current positioning centers on production silicon: the AML100 shipping and in design-in, and the AML200 in development, applied to counter-drone detection (SkyML), wideband RF sensing and classification, industrial IoT anomaly detection and always-on security. Its most recently announced financing is the $5 million Series B closed in September 2023, which included new strategic investor and partner Unitrontech for automotive AI applications.