Any form of product design – be it an IC, a car or a bridge – primarily has to deliver the required functionality. But it is the myriad of boundary conditions that makes the design puzzle really interesting. A bridge connects the two shores of a river, but it truly becomes a challenge to build one at a height of 342 meters, as is the case for the Millau Viaduct.
Every designer has to maneuver through his own application-specific limitations and find original solutions that avoid undesired behavior while strengthening the demanded performance. Electronics designers combine components in a semiconductor technology into a relevant function. They’re confronted with the specifications defined by the application, but also need to adapt to the technological limits such as trading off speed gain versus the continuous reduction of power supply voltages.
For more than half a century, the only tool that designers of PCBs and integrated circuits were using is a simulator in combination with parasitic extractors. Mastering designs at the limits of the technological boundaries with little CAD support is not simple and requires years of experience. It’s not surprising that successful designers of advanced electronic circuits experience a certain reverence.
Saving just one design iteration can turn a market introduction into a huge success. A National Semiconductor manager was once asked why he didn’t hire fresh engineers from universities but kidnapped experienced and expensive designers from the competition. His answer was: “Let the other companies pay for the failures of young engineers.”
Obviously, design automation companies see in this struggle a potential market: what do designers need more than simulators and extractors? Or even more aggressively: how can we make these elusive analog artists superfluous?
When CMOS technology became widely available for both digital and analog design, the difference in design productivity between both became apparent. Attempts were made to automate parts of the analog design trajectory by relating the desired specification parameters to the design choices through a large set of equations. Unfortunately, when the tooling was finally tuned to a particular technology node, the industry was already moving to the next.
Many more efforts (eg Neolinear, Opmaxx, ADA) were made to automate the analog design process. Barcelona Design attracted a lot of attention with a web-based tool. In contrast to a digital library cell, which is copied in hundreds of designs, an analog circuit is never exactly reusable in another product. As a consequence, the amount of required support blew their business model to pieces.
Solido Design and MunEDA took another route: let the designer propose the circuit topology and the CAD tool will tweak the circuit towards the optimum performance corner. As one of the fastest growing US companies, Solido was acquired in 2017 by Mentor.
The newest kid on the block is Agile Analog. Headed by former ARM managers, the Cambridge-based company claims to deliver quicker, more accurate and cheaper designs. The holy grail to make this promise come true nowadays is obvious: artificial intelligence. However, a closer look at their product portfolio reveals designs “based on traditional architectures”. Most likely, their success will be more dependent on the quality of their design staff than on this mythical AI design philosophy.
An AI approach to design requires to formulate all these slippery specification requirements into unambiguous software code. How to translate into code the requirement that the circuit must operate near the 7 tesla field of an MRI scanner or at 4 kelvin for a quantum computer? Or the possibility that a blocker signal appears at an arbitrary node in the circuit, or degradation and reliability issues?
Experienced designers make an educated guess in what design phase each specification has to be considered, and where the electronics design gradually moves towards finding trade-offs on other product levels. An optimal electronics design always interacts with its intended environment. This delicate translation process of requirements into software remains the true bottleneck to analog design automation.