Marcel Pelgrom

Marcel Pelgrom consults on analog IC design.

24 August 2022

Will there ever be a technology that can compare to silicon integration? Marcel Pelgrom hopes so.

Some 40 years ago, I attended a lecture by a veteran professor, a true teacher. Where today most Powerpoint fetishists use advanced features to retain the visual attention of the audience, he just drew a horizontal line on the blackboard. The expectation bar, he called it. His talk was about which technological developments had exceeded his expectations and which he considered to be underperforming.

In the last category, a few themes appeared, like space exploration. Ten years before, the astronauts had visited the moon, but what had been the benefit to humanity? Another one was the perceptron: an attempt to mimic human intelligence in a machine. The only noteworthy thing that came from that was the “2001, a space odyssey” movie (1968).

Somewhat closer to the expectation bar were themes like magnetic storage, display technology and sensors and actuators: useful technologies with important application domains. However, these technologies aren’t platforms, from which many new technologies can be developed for a variety of different markets.

Of course, such a platform had come into existence and held an indisputable position far above the expectation bar: silicon integration enabled enormous computation power, serving a whole range of software applications. Moore’s Law unified the entire engineering supply chain towards one goal: to put more functionality on one chip. Materials engineers, technologists and IC designers shared a challenge and thereby managed to beat even the most extreme predictions.


Free webinar ‘Modernizing your code base with C++20’

As many production tool chains now adopt C++20 features, the potential this brings is unlocked. What advantages can recent versions offer to your code base? In this webinar we’ll look at the great improvements C++ has gone through and how features like concepts and ranges can transform your code. Register for 2 February, 4PM.

How is the expectation bar looking today? For me, good ideas go fast and pay for themselves. The US telephone patent was filed in 1876 and five years later, the telephone exchange in Amsterdam became operational. Its 21st-century equivalent is the smartphone. One of its enabling technologies, ie wireless technology, has exceeded all expectations: in introduction speed and volume. Applications via the internet and social media erupted within a few years.

Many reasons were brought forward why Moore’s Law would flatten: economics of 10-billion-dollar plants, too much power consumption, the communication bandwidth with the peripherals and the gap between what technology would offer and what designers could build are just some of the arguments. Still, silicon integration dominates technological development.

In its slipstream, other media changed as well: flat thin-film transistor displays replaced CRTs and optical storage grabbed part of the magnetic storage market until it was replaced by … silicon.

Various alternative technologies were pursued to replace silicon. III-V materials remain a niche. Polymers allow computing on flexible sheets at low performance and are found in polymer OLED displays. Adiabatic computing, carbon nanotubes and nanotechnology didn’t really fly. Multilevel logic has been confined to memories. Photonics has been a hype for four decades. It will stay that way as fundamental functions – a pure optical switch or an optical buffer – are lacking. Similarly, quantum computing with its probabilistic results at zero kelvin is a ‘never-ending promise.’ It’s really disappointing that photonics and quantum technology are considered key technologies by the Dutch government, the Dutch Research Council (NWO) and other organizations.

Ubiquitous computing, pervasive computing and ambient intelligence have aimed at a closer interaction between humans and computational power. Some of these techniques are in use in a limited number of applications, but their impact can’t compete with the mobile-phone revolution. And remember neural networks? A failed attempt to mimic human intelligence in a machine.

What will the expectation bar look like in the coming decades? Any fundamentally new technology faces the legacy problem: what idea can make the gigantic investments in silicon hardware and software obsolete? Therefore, silicon technology will expand further, eg in three dimensions as today in memory technology. Ultimately, the human brain is also three-dimensional, and moreover self-organizing, self-learning, self-correcting and adaptive. Room for silicon to improve. Artificial intelligence attempts to mimic human intelligence through pattern recognition. Its one-dimensional nature and associated limitations proved this technology still needs several fundamental breakthroughs.

Hopefully, the top position of the future expectation bar is the solution to the main challenges ahead: climate change, energy storage and pollution. Maybe in two decades, people will reflect and conclude that the solution to these challenges has a parallel to Moore’s Law: if humanity manages to focus all its engineering attention on a specific topic, we can beat our wildest expectations!

Marcel Pelgrom has been writing columns for Bits&Chips since 2002. This is his last contribution. His wit, his candor and most of all his provocative insights will be sorely missed!