Reading time ( words)
Shaughnessy: You know more than anyone that most designers don’t use an autorouter, or they won’t admit it, because they have a beauty contest with their co-workers. Many designers would rather work an extra day just to have the traces look pretty on both sides of the board.
Liao: Autorouting still isn’t easy, but we tried to figure out a way to automate it as much as possible. The purpose is not how much you can automate, but how much productivity can be gained by utilizing our tool.
Shaughnessy: Is it a mixture of auto-assisted and fully automatic, where you can let it run, and then take over and do some things on your own?
Hanzelka: Both are options.
Liao: For some complex, difficult patterns, the user should be doing some manual routing and then let the router finish the rest of it, which still saves time. Also, we try to provide some high-level interaction for the user so that they can define a global path, such as some bundles. Sometimes, you just want to use it to define a single avoidance path; that’s very difficult to achieve by default, but we can let the user define this global guidance or path that makes the router follow the guidance and still make a good routing design. Again, our goal is to make the router as high-quality as people want—not just for the autorouting.
Hanzelka: Our goal is to provide autorouting results to be as close as possible to hand-routing.
Shaughnessy: How much does AI or machine learning play into this?
Liao: Traditional place-and-route technology is one of the classic problems with AI, so this problem is already in AI. But what usually people are currently talking about with AI is machine learning. We also use that kind of methodology to make the result better. We are not doing something that is generic AI because the thing is very difficult. The goal is possible because it’s limited, such as a 19x19 area, but you cannot learn all of the PCB designs; that would be very difficult. It’s a very interesting area. All of the routing technology is optimization technology, which was the power of the AI area for a long time. We are doing that and we also call it specialized AI.
Shaughnessy: That sounds great. We were joking among ourselves a couple of weeks ago that there hadn’t seemed to be any big advances in routing since CCT.
Liao: Right. Personally, I divide routing technology into three generations. The first one is gridded routing. The second is gridless, or shape-based routing, and CCT was one of the pioneers for shape-based routing, which has become very popular, especially in PCB design. I would say the third generation is topological routing. So, we are in a new generation of routing technology. And the good thing about topological routing is that it’s very flexible and can handle almost any routing. It was very difficult before because of any angle in routing. Also, topological routing is much faster compared to shape-based routing.
Recently, more and more companies are serious about doing the whole board with autorouting, but customers still do benchmarks where the completion rate is 98% versus 70%. I think it’s mainly because the topological routing is really thorough and a completely different approach; for some designs, it’s difficult to achieve high completion, but it’s possible now.
Shaughnessy: Years ago, designers would say that the autorouter would get 80% of nets done, but fixing the remaining 20% would take 80% of the time. The 80/20 flipped on its head.
Hanzelka: One of our latest endeavors is push-and-shove tuning. You can take a group of routes which are already completed and totally embedded within other routes, go in, adjust the length of them, and tell it to auto-tune them. Then, our router will commence pushing and shoving using trombone or serpentine patterns in order to get the additional required length. There is no need to delete the traces, reroute them, and do a large amount of clean up anymore. Now, you can just push a button, let it run, and it’s done.
Shaughnessy: That’s cool. Where are you based out of now?
Liao: The Bay Area.
Shaughnessy: Is there anything else you want to add?
Liao: In the 20th century, a large number of companies were using CCT routers, but now fewer people are seriously using autorouters. The time has come again to utilize autorouting tools as much as possible and significantly decrease the design time shorten significantly. Our autorouter has the technology to eliminate the need for manually cleaning up the routing.
Shaughnessy: It did always strike me as crazy that there was this fairly clever tool, but a lot of designers were not using it.
Hanzelka: Well, the problem was that you’d route it, and then have to go back in and reroute numerous sections due to poor routing paths, or clean up the errors introduced by the autorouter. But we don’t take that approach; we want to do it correctly the first time.
Liao: Also, I mentioned we achieve a high completion rate, but that’s just one of the aspects of this technology. Our goal is to create a route result that is exactly what the user wants. You can achieve high completion rates, but the user still needs to modify a lot, which is why we design some interactive features to let users decide what sort of outcome they get.
Shaughnessy: So, they have more control. I think that’s the thing they don’t like is giving up control.
Liao: Right. We try to give more control to the users.
Hanzelka: And if they’re getting the results that they would normally get from hand routing out of the autorouter, then I think they’re willing to give up some control for efficiency.
Shaughnessy: Thank you. We’ll talk again soon.
Liao: Thank you, Andy.