Richard Hegberg is joining US ultra low power analog AI processor designer Aspinity as chief executive.
Hegberg joins as the Aspinity AML100 analog AI processor is entering volume production aiming at the Internet of Things (IoT) and automotive applications. The company has deals with Infineon Technolgies, Renesas Electronics and Korean automotive chip distributor Unitrontech.
Hegberg brings more than 25 years of executive leadership experience in semiconductor technology from Vesper Mems, NetApp, SanDisk and Qualcomm. He replaces co-founder Tom Doyle, who served as Aspinity’s CEO for the last 7 years, will continue to support the company’s growth as a board member and takes on the role of Chief Operating Officer.
- Infineon gets ‘always-on’ with Aspinity deal
- Aspinity raises funds for analog ML roadmap, expansion
- Renesas-compatible machine learning dev board
The AML100 uses an analog neural network called Reconfigurable Analog Modular Processor (RAMP) for a low power, always on processor. This uses parallel, independent analog circuit blocks that operate in the subthreshold domain. Each block is implemented in a very small footprint and each can be independently powered only when they are needed for a specific task.
The RAMP technology also uses non-linear analog circuitry to improve the performance of typical analog tasks, make decisions, and classify incoming sensor information.
Aspinity has patented a 10bit analog non-volatile memory (NVM) that is implemented in standard CMOS with no add-ons and sits alongside the analog computing elements in each analog block. This can be used to both store neural network weights as well as biases and activations for other compute circuits. It can also store the high precision values needed to finely trim out variations in analog circuit performance that arise from environmental conditions or the CMOS manufacturing process.
The complete functionality of a RAMP-based chip can be abstracted from hardware into software, enabling a flexible analog platform in which all aspects of the chip (connections of circuit blocks, parameters, etc.) can be programmed in software and stored in on-chip memory.
“It has become clear to me that Aspinity’s technology and IP portfolio delivers unique sensing capabilities at near-zero power, utilizing the analog ML processor. The company has demonstrated this value in a variety of proof of concepts across automotive, dash cam, IOT, security, and data center applications,” said Hegberg.
- Analog AI startup gains Korean automotive distributor
- Analog machine learning chip lowers ‘always-on’ power
“Aspinity has invented a foundational all-analog neural processing core that is transforming how real-world data is processed. In addition, the core’s all-analog neural accelerator, which can deliver industry-leading performance and efficiency, will be an invaluable IP core for many companies looking to build neural processors for AI applications. I look forward to helping the company scale while driving commercial success.”
In the automotive industry the company has demonstrated the ability to distinguish critical security and condition-monitoring events, such as damage, collisions, and glass break from hundreds of disturber events — all while consuming less than 20 microamps of supply current. For automotive dash cams and surveillance systems, Aspinity’s technology will allow next-generation AI-driven dash cams to operate in an always-on, parked surveillance mode while consuming near-zero system power.
“Richard Hegberg’s track record in elevating companies with unique breakthrough technologies to billion-dollar revenues was a decisive factor in his selection,” said Tom Doyle. “With his leadership, Aspinity is poised for significant growth in product development and IP, cementing its position as a leader in all-analog ML/AI processing.”