Very soon we’ll start a new series looking at the story of the development of chips dedicated to accelerating machine learning, with an initial post looking at the origins of Google’s first TPU.

But Google’s TPU development, starting in 2013, wasn’t the first project to create dedicated neural network chips. Intel, amongst many others, got there first. Decades earlier, in fact.

And reporting on Intel’s first effort was columnist, now turned podcast host, John C. Dvorak. Writing in PC Magazine:

While everyone worries about the next hot processor chip, a small team of engineers at Intel invented the 80170, the world's fastest and most powerful neural network on a chip. In processing power, it is something of a breakthrough, having achieved the theoretical intelligence level of a cockroach - no kidding.

I visited with the chip's developers, who admit that this analog computer possesses a frightening amount of processing power and has the company baffled. “Nobody at Intel knows what to make of this thing," I was told.

Neural nets aren't programmed — they're trained — like a dog. The programmer's job (if one can be called a programmer) is to organize the data in such a way that it can be analyzed over and over until the network discovers patterns and extrapolates data from incomplete data sets. Predicting weather and determining which horse will win the third at Belmont are perfect neural net applications.

But after the humor, and perhaps a degree of scepticism, there is a prediction that would take decades to come true.

Don't be left out. There's something big lurking beneath the surface of this technology.

When did the Intel 80170 appear? All the way back in 1989.

To put this into context, this is the year that Intel’s 80486 first appeared. A typical high end PC had an 80386 running at up to 20MHz and up to 16MB of memory.

The 80170 was quite different from these other Intel products though. It was an analog neural processor. According to Wikichip:

ETANN (Electronically Trainable Analog Neural Network) was one of the first commercial neural processors, introduced by Intel around 1989. Implemented on a 1.0 µm process, this chip incorporated 64 analog neurons and 10,240 analog synapses. The ETANN is also the first commercial analog neural processor and is considered to be the first successful commercial neural network chip.

Intel 80170 block Diagram

We’ll look in more detail at the 80170, and other early neural network chips, in later posts. In the meantime, well done to John Dvorak on his prescience.

A brief footnote. I discovered today that Dvorak is the nephew of August Dvorak, who invented the Dvorak keyboard layout.

If you’ve enjoyed this then you might like:

Neural Networks and the Origins of Modern Computing

Babbage

September 17, 2023

In today’s post we look at how some of the very earliest work on neural networks helped influence the development of the modern ‘stored program’ computer. I got attached to this McCulloch and somewhere I wrote, maybe 100 years from now he will be seen as the greatest 20th century philosopher along with Russell and a few people like that.

Read full story