Caveat Lector
Portrait photo of a smiling Mike Buetow
mike
buetow
editor-
in-chief
Googling Your Boards
I

f the market is big enough, sooner or later Google will join it.

That much was laid bare in late June when the search giant cum OEM announced its latest venture, Visual Inspection AI, a new “purpose-built solution” designed to help businesses, including manufacturers, reduce defects and cut operational costs.

Now before you start doubting Google’s temerity to dive into technology that cuts across almost every industry imaginable, remember we’ve been here before.

While the company today still counts on its hugely successful targeted search marketing program for the bulk of its revenue and profits, several other businesses it has launched have made serious inroads in their respective markets. These include broadband; telecommunications; autonomous vehicles; and human health gambits (marketed under the Verily Life Sciences name). Acquisitions brought it Nest Labs, the maker of smart thermostats. Less front and center, but just as integral, are Google’s vast data centers, also known as server farms, which power its reach into just about every precipice known to man.

In some respects, Google is starting to look like the industry-conquering Japanese OEMs of the 1980s, or perhaps Siemens or Philips, albeit with significantly greater profits. As it closes the gap with other conglomerates, electronics was almost certainly going to become part of the mix.

Artificial intelligence is the hot buzzword. Going back a few years, we have seen many suppliers to the electronics manufacturing industry adopt the phrase for their tools, deservedly or not. Besides the major OEMs, third parties have entered the field, attempting to tackle entrenched and labor-intensive problems. But the problem all these companies run into is that, while their tools can handle simpler components or products, there’s simply too many complex ones to make it a panacea.

Google doesn’t come to this completely oblivious. Its AutoML product is said to enable developers, even ones with limited machine learning expertise, to train high-quality models specific to their business needs, but that solution to manufacturing quality control suits general purpose needs. Visual Inspection AI is Google’s answer to manufacturers that need specific solutions.

The platform is based on Google Cloud’s computer vision technology, and according to the company can build accurate models with up to 300 times fewer human-labeled images than general-purpose ML platforms. Keep in mind how relatively simple most AOIs are to operate today. For Google to attain that kind of startup speed is essential to quick, easy deployment, which is a must if it will compete with the current infrastructure of machines.

The rest of the fine print is less fine, however. Google’s beta tests were run against general-purpose ML machines, which isn’t the way to baseline its effectiveness in an electronics manufacturing environment. And the company touts – ironically – that its deep learning tool allows customers to train models that detect, classify, and precisely locate multiple defect types in a single image. That manual intervention is exactly what manufacturers hope to eliminate.

Through all of this, Google promotes that the AI rests in its Cloud. I suspect what Google hopes is through large numbers of users, perhaps from scores of different companies, it can build a database by which its “AI” will recognize common defects (all while Google collects fees from the subscribers). We’ve reached out to Google for more details, with no response as yet.

FIH Mobile, a subsidiary of Foxconn, is one of the beta sites. Foxconn is notoriously quiet, but the fact that it permitted Google to use its name in the announcement may suggest we’ll get a look at their data down the road. We won’t hold our breath.

There’s no question AOI must continue to improve. Google’s decision to try to do so is fascinating. But from what I see, this time might be a no-go.

Mike Buetow
P.S. Those truly interested in knowing what their inspection system is saying should tune in for David Bernard’s webinar on Sept. 9 (pcb2day.com). And if you are missing human connections, register for PCB West, coming October to the Santa Clara Convention Center (pcbwest.com).