Black magic — Things that work

21 April 2023

Most friends know that my first career was semiconductor. Few were familiar enough with the industry to know that specifically I was in this very special domain called “analog IC design”.

Most friends know that my first career was semiconductor. Few were familiar enough with the industry to know that specifically I was in this very special domain called “analog IC design”.

I started working on analog IC design back in my senior year in college (1997), when I signed up on a research project with Prof. Shen-Iuan Liu, who would later become my master thesis advisor. (Prof. Liu would be named IEEE Fellow a few years after I graduated.)

Upon finishing my master degree in 2000, I embarked on what would be a 12-year career in analog IC design (among which 4 years as an entrepreneur) that would take me from Taiwan to Silicon Valley, to work on and ship hundreds of millions of chips that contain critical analog technologies in WiFi, Bluetooth, USB, Ethernet, HDMI, Displayport, etc…

An example of analog design (from Homebrew RF Circuit Design Ideas)

What people outside of the industry don’t know was that analog IC design is sometimes compared to “black magic”, and not in a joking way.

Unlike digital design where the written design codes based on languages like Verilog or VHDL translate into final designs that almost always work as designed, analog design is this frustrating exercise of heuristic for most junior and senior practitioners.

That’s not to say analog design is pure trial-and-error. Far from it.

All analog designs are built on mathematical equations that could be as high-level as S-domain equations or transistor-level Kirchhoff current/voltage equations. Sometimes we get as low-level as to electrons and even microwave derivative equations. While the digital/logic designers had long shifted into pure computer-assisted works, analog designers remain a rare animal that still carry a lot of paper-and-pencil calculations in parallel with the sophisticated computer-aided simulations run on HSPICE or other analog simulation CAD tools.

An example of S-domain transfer function

Analog design is without doubt among the closest to theoretical science (physics/chemistry) in the entire semiconductor industry — probably second only to foundry process technology developments like those carried out by TSMC. Samsung and Intel.

If you ask a digital/logic designer to explain to you why the transistor radio your grandpa left you was generating so much noise recently, he would have trouble even coming up with a remotely sensible explanation. But a seasoned analog designer would likely be able to take a look at it, do some test and give you some reasonable causes.

Heck, give him a soldering gun he might even fix it for you!

So if it’s so close to science, why is it compared to black magic as I mentioned earlier?

Analog design is sometimes compared to black magic

In fact the reason is actually exactly because it’s closest to science, it’s also prone to all the kinks that could not be explained by science or its army of beautiful closed-form equations.

For comparison, if you are a programmer, debugging means that your codes don’t compile and run as you intend them to do so and you’re trying to figure out what you did wrong by plowing through thousands or tens of thousands of lines of codes. Chances are that you will always find out what went wrong, as long as you have a very strong sense of logic and understand how to systematically weed out the bug. We often hear about stories of programmers scratching their heads through sleepless nights in the office trying to find the bug. We seldom hear stories where the software bugs remain missing despite all the effort.

The same thing with digital design. The simulations (assuming properly set up) always correspond nicely to how the eventual chips function. Performance such as speed might not be as simulated — remember those days where Intel tranched the CPUs into grades of different max clock rates and sold at different prices? — but function-wise it’s always as intended.

This is not the case in analog design.

I had never in my 12-year career seen any actual chip that yielded the exact same function and performance as designed and simulated. What’s even worse, many bugs remain unresolved despite weeks or months sleeping in the labs and stinking like hell.

In fact, I had a circuitry that I designed in grad school, which 12 years later when I left the industry I still didn’t know why it didn’t work.

Of course when we debugged analog designs we relied heavily on science and scientific equations. It’s very common to walk into an analog labs and found a dirty white board scribbled with all kinds of derivative or S-domain equations. Same on those small white boards in the cubicles of me and my former colleagues in the Silicon Valley.

But theories only give us directions and get us close. Sometimes however you try, things just don’t work.

When you’re a junior designer you’d be so angry — “Why doesn’t it work? This doesn’t make sense!”

If you’re not smart enough you might blame it on process technologies (note to self: it’s rarely their fault) and ask to respin the same design.

Even if you get what you ask for, 10 out of 10 times you’ll find the same results. What doesn’t work remains not working, despite all your theories and calculations saying that it should work.

On the other hand, what’s proven to work tends to keep working, even if the theory says that it shouldn’t work.

All analog designers of more than 10 years of experiences have a few tricks up their sleeves that when asked about, they would just shrug and say “It works.” These tricks would not be able to make it to engineering journals such as IEEE Journal of Solid-State Circuits (JSSC), or into presentation of conferences such as International Solid-State Circuit Conference (ISSCC).

But they work. And there are many of them in commercial products — in your iPhones, your Macbooks, your digital cameras, your LCD TVs, etc.

This is the №1 lesson for an analog designer: Forget about the theories. Stick with things that work and ditch things that don’t work.

So this is why people in the industry sometimes call analog design “Black Magic” — though it’s uncommon that they call us magicians or wizards (which would have been really cool!)

A side-effect of this black magic characteristic is that analog designers take more than 10 years to formulate and the really great ones ended up being a rare breed — as rare as the Witchers.

I often say semiconductors are the closest thing to magic in the modern world. Analog is like the black magic because there are only a small number of people in the world today who can design these analog chips. It’s trial and error, it takes a very long time, and it’s almost as much of an art as it is a science.

Gavin Baker, founder and Chief Investment Officer of Atreides Management

Above is from a recent interview of Gavin Baker, one of the few famous investors today that still knows semiconductor industry inside out.

For him, he cares about only whether he can make money by investing in semiconductor stocks like those of Nvidia or Intel. He doesn’t care whether there was black magic behind it that could not be explained by science.

The same should be for you my dear readers. You should not care whether there is black magic involved in your iPhones, Macbooks, digital cameras, LCD TVs, etc. You should only care whether they work or not.

If you insist on a magic-free world entirely run on science, you might as well go back to life with only papers and pencils. There is much less black magic behind them.