Issue link: https://iconnect007.uberflip.com/i/929192
30 DESIGN007 MAGAZINE I JANUARY 2018 is applied real-time to the BOM document in the design project, and warnings and errors are shown automatically for each line item. In the example shown here, the selected component is obsolete, has outdated supplier links and no user preference rank for any given supplier. This lets the user know they need to find a drop-in alternative to this 56pF 0402 capacitor, which would not require a physical change to the PCB itself. Of course, sometimes the component that is no longer available is more complex, like a microcontroller, and it would trigger a new revision of the PCB to replace with an alterna- tive that's not a drop-in replacement. However, the BOM rule checking and live supply chain links make this scenario far less likely. Deep Learning (Applied Artificial Intelligence) PCB designers and engineers who I talk to (including myself) will always say the same thing: "An autorouter makes a mess and can- not do what I can do. I will never use an autorouter!" And fair enough, too! For all but the simplest designs that have no impedance control, no sensitive analogue circuits and no high-speed topology requirements, our experi- ence of routing automation in times past has been less than stellar to put it mildly. From my own point of view, what's around the corner with AI and deep learning is the crux of this matter. With all those complex boards to design, we need automation to help with component placement and routing of the PCB. We need it whether we want it or not. But it must work. I mention AI and deep learning because we are finally at a point in time where we have technology that can rapidly accelerate the learning of routing and placement algorithms. Yes, it's been tried several times before. Neural nets, for example, have been used extensively in the development of 1990s era autorouting tools, such as Neuroroute. Just a layman's recap here: A neural net is a connectivity or relationship graph that is "trained" by first feeding in examples of what it should do with human generated results. The idea is that the neural net "learns" the human way of rout - ing PCBs and creates a set of parameters that cause it to mimic how a human would do the routing. Neuroroute (Figure 4) was arguably the first ever neural-net based topological autorouter ever developed. It was acquired by Protel (now Altium) in the late 1990s and formed the base engine of what became the Situs autorouter that is in all Altium PCB tools to this day. I say it had great promise, but for early-'90s PCB designs (two-to-four layers, 15 mil track-space ratio, 1 mm pitch QFP and 0805 chip compo - nents), it actually did a pretty decent job. The problem is that it was not able to adapt to the much finer package pitches and BGAs that hit PCB designers in the early 2000s, and the basic set of PCBs used to "train" it was very limited. The principle was sound, but lack of ongo- ing training data set availability makes it hard to update. Since then, many largely improved routing tools have been developed. The topo - logical router can be compared with tradi- tional shape-based routers (Figure 5) such as SPECCTRA by mapping the board space and obstacles using polygons, and applying differ - ent path costs (for example, adding vias ver- sus going the long way around the board), to Figure 4: Neuroroute user manual front page (Source: Altium LLC 2017).