Why analog design resisted automation for forty years

Digital design was transformed by synthesis and place-and-route in the 1980s. Analog never got its revolution — and the reasons why are exactly what makes this moment different.

Ask any chip designer why analog is hard and you'll get the same answer, phrased a hundred ways: there is no clean abstraction between what you want and the transistors that deliver it. Digital design broke that dependency decades ago. A designer describes behaviour in a hardware description language, and a chain of automated tools compiles it down to silicon. Productivity exploded. Analog was left behind.

Continuous problems don't quantise neatly

Digital logic deals in discrete states — ones and zeros — which makes it tractable to abstract, optimise, and verify with rules. Analog deals in continuous quantities: voltage, current, noise, temperature. Every parameter interacts with every other. Nudge a transistor's width to improve gain and you may have just wrecked your noise figure, your matching, or your stability. There is no layer of abstraction that hides this from the designer.

The knowledge is tacit

The hardest part of analog design isn't the mathematics — it's the judgement. Which topology to reach for. How to size a bias network. Which parasitic will bite you at layout. This intuition is accumulated over decades and lives in the heads of a shrinking pool of experts. It was never written down in a form a rules-based tool could execute, which is precisely why traditional EDA never automated it.

Every process is its own world

Analog expertise doesn't transfer cleanly between foundries. Each process has its own device physics, its own quirks, its own Process Design Kit — and that PDK is almost always proprietary, locked behind foundry NDAs rather than openly accessible for study or generalisation. A designer fluent in one node can be a novice in another. That fragmentation multiplied the specialist knowledge required and slowed every attempt at generalisation.

In recent years, however, a meaningful shift has begun. Through the diligent work of industry and academic advocates, a number of advanced node PDKs have been made accessible to the open-source community — a significant step toward democratising hardware design and lowering the barriers to entry for researchers, startups, and independent engineers.

Why now

The thing that resisted formalisation — tacit, example-driven intuition — is exactly what modern AI is good at learning. Rather than hand-coding rules, we can learn from verified designs, capturing the patterns that expert engineers apply without being able to fully articulate them. That is the unlock. The automation gap that stood for forty years is closing not because analog got easier, but because the tools finally match the shape of the problem.

— The Samurai Team