Estimated reading time: 2 minutes

The Case for Artificial Intelligence in EDA Tools

I-Connect007 Editor Andy Shaughnessy reported that the keynote speaker at the IPC APEX EXPO Design Forum was Dean Parker, a former PCB designer at Shure who is now a CAD manager at Google X. Parker is involved in the development of autonomous vehicles and all sorts of other great ideas at Google X. According to Andy, Parker told the crowd, among other things, that EDA tool vendors need to trash all their old 1990s code and start over, this time with artificial intelligence.

There has been a lot of activity in the field of AI recently, with such developments as voice recognition, unmanned autonomous vehicles and data mining to list a few. But how could AI possibly influence the PCB design process? This month, I will take a look at the endless possibilities.

So much time is wasted on reproducing the same thing over and over again on each layout. Current EDA tools, with all their bells and whistles, are still very limited in automation processes and mostly rely on the skills and foresight of the engineer and PCB designer to drive the software through all the hoops. Instead, EDA tools need to predict what the designer is trying to do, then look at previous designs to suggest alternatives and auto-complete the design where possible. AI is a system that perceives its environment and takes actions to maximize its chances of success.

Automating many of the tedious steps in setting up the initial database would be a good start. A standard form factor could be used to establish the initial layout environment ensuring that designs are compatible across multiple generations of technology. Although some PCB layout tools allow the designer to load a standard set of predefined startup configuration files, there is still too much manual intervention required. The PCB database could predict the fundamental design rules and via stack requirements sourced from previous experience.

Predictive text, which we all use every day on our cell phones, could provide self-evident naming conventions for supplier part numbers and database fields, greatly speeding up the design definition. Busses and interfaces could be analyzed and categorized with naming conventions interpreted from the chip pin name assignments, eliminating much of the monotonous schematic capture process. IC power pins could have powers supplies assigned based on datasheet requirements. And a starter set of decoupling capacitors, added to each power pin, could kick off the PDN analysis based on previous capacitor availability and parameter selections.

A selection of predefined library components could be offered, based on an initial bill of materials, and pre-placed on the schematic predicting the designer’s requirements. IBIS models could be automatically assigned to each chip, based on the part number and all the interconnecting transmission lines identified. The IBIS model’s source and load impedances could be extracted to assign the required impedance and terminations to each individual transmission line.

Also from this, the board stackup could be created based on previous designs, with similar technology, selecting dielectric materials, from a well maintained library, sourced from the preferred fabricator availability, dielectric loss and bandwidth requirements. Data and address busses together with clock/strobe different pairs, defined at the schematic entry level, could be assigned to certain layers in order to minimize crosstalk, electromagnetic emissions and return path loops. Power plane shapes could be automatically defined based on component placement and on the pins that need to be connected, allowing for DC drop and maximum current supply.

To read this entire article, which appeared in the May 2016 issue of The PCB Design Magazine, click here.