Launches "Effortless ML" MLSoC Chip, Evaluation Board for "10x" Edge AI Performance

Offering 50 TOPS in a 5W envelope and with a dedicated computer vision coprocessor, the MLSoC aims to blow the competition out of the water.

Machine learning specialist claims to be delivering on its goal of a tenfold boost in performance per watt for computer vision at the edge with its freshly-launched Machine Learning System-on-Chip (MLSoC) — the company's very first silicon, designed to deliver what it calls "Effortless ML."

“When we started 3.5 years ago, we set out to deliver a disruptive 10x performance improvement over alternatives and provide a scalable industry-leading ML experience solving computer vision applications," claims company founder and chief executive Krishna Rangasayee. "Today we are delighting customers by delivering on that promise and exceeding their expectations. We are excited to take our very first purpose-built software-centric MLSoC to volume production." has officially launched the MLSoC, a "software-centric" machine learning chip targeting the edge. (📹:

The MLSoC is built on TSMC's 16nm fabrication node, but the hardware takes a backseat to the company's focus on software — through which it promises to offer ten times the performance of its edge-processing competition for any computer vision application, regardless of sensor type, framework, neural network, or model.

According to, these bold claims are delivered thanks to a novel architecture in its machine learning accelerator capable of pushing 50 trillion operations per second (TOPS) in a 5W power envelope. The system-on-chip design also includes a quad-core Arm Cortex-A65 application processor running at 1.15GHz, an Synopsys ARC EV74 vision processor, hardware H.264/H.265 decode at 4k60 and H.264 encode at 4k30, 4MB of on-chip memory, 32-bit four-channel LPDDR4 memory controller, eight PCI Express Gen. 4 lanes, four gigabit Ethernet ports, SDIO and eMMC, two SPI8 buses, two I2C buses, and 32 general-purpose input/output (GPIO) pins.

"We partnered with TSMC to manufacture our MLSoC Platform because we wanted to work with the best," claims's Gopal Hegde. "Our first-time-right silicon success allowed us to begin shipping products to customers immediately. Our customers have been waiting for a purpose-built software and hardware MLSoC platform and we are thrilled to work with TSMC to deliver the industry’s first and only solution that enables Effortless ML."

In addition to the chip itself, has launched an evaluation board which places the MLSoC on a PCI Express card with 16GB of RAM and 16GB eMMC storage — operable as an accelerator card in a host machine or stand-alone single-board computer with four gigabit Ethernet ports for connectivity to external systems.

More information on the MLSoC is available on the website, though the company is not yet providing public pricing for the SoC nor the evaluation board - asking interested parties to fill in a form on the website for more details.

Gareth Halfacree
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire:
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles