This Month | The Ojo-Yoshida Report
Skip to content

This Month

Hardening 'Brittle' AI

Moving Beyond 'Brittle' AI

brittle AI

By George Leopold

What’s at stake:
Most machine learning models developed for corporate applications never make it to production. Greater engineering discipline and standards are needed to catch failures earlier in the development process.


AI Hardware Startups Ready for Pruning

pruning AI startups

By Peter Clarke

What’s at stake:
With plenty of VC funding on offer in recent years to bankroll an AI gambit, entrepreneurial engineers have been only too happy to accept the cash and assume the risk. But where are the returns?

There are too many artificial intelligence and machine learning startups. The going is getting tougher. Consolidation and acquisitions are bound to follow.

Just how rocky the AI market has become is illustrated by early mover and well-funded startup Graphcore Ltd., which announced layoffs last week. The announcement follows similar cuts at Mythic AI made earlier in the summer. If these established AI pioneers are axing jobs, what are the prospects for the legion of smaller AI wannabees?

We’ve seen about a decade of development of hardware implementations for neural network acceleration and the resulting AI algorithms. There are now probably more than 100 hardware-oriented startups still active and trying to cash in on the biggest revolution in computation since the adoption of the von Neumann architecture in the 1950s.

The accompanying roster of AI startups, organized by founding year, lists 90 entrants. All are fabless, using foundries to manufacture chips, and many are incorporating AI architectures in FPGAs or looking to license designs as intellectual property. Usually in such domains, the software- and services-oriented startups dwarf the hardware cohort by a factor of 10 or 20 to 1.


Who's Who in AI Hardware Startups

AI technology startup

By Peter Clarke

Of those 90 startups we have identified, many will falter, others will be acquired. A handful may survive. Our analysis is here. Below is our list:


Inside In-Memory Computing, and Why It's Back

in-memory computing and AI

By Ron Wilson

What’s at stake?
In-memory computing, an old and controversial way of organizing computer hardware to minimize energy consumption and maximize performance, has never quite broken through into the mainstream, except in some very specific applications. But the needs of edge-computing AI may provide an opportunity for a unique embodiment of this architectural idea.


Can AI in AVs Go Beyond 'Perception'?

vehicle electronic control

By Junko Yoshida

What’s at stake?
Artificial intelligence is commonly used in providing the perception abilities to highly automated vehicles. Can AI also help AVs make safer decisions in planning and control? Anyone accustomed to deterministic algorithms based on control theories will be hard to convince. But Infineon has a radical rebuttal and has been arguing its safety case with stakeholders in the automotive industry.