The Printed Circuit Board Market is an important albeit a small part of the overall EDA market. The table below shows the size of the PCB and MCM market for the last two years.
Source EDA Consortium
The combined PCB and MCM revenue accounts for about 7.5% of total EDA revenue. This market is flat. The market is dominated by Mentor Graphics, Cadence and Zuken. I had an opportunity to interview representative of two of these firms.
I spoke with David Weins Director of Business Development before Mentor's April 10th announcement of XtremeAR, its new autorouting product for large PCB boards.
There are two different methodologies that have been used over time for PCB design. There is the automated approach which you would think everybody would want to do. Then there is the manual approach. What we continue to see is that there are certain designs that kind of defy the automation that we would like to see everybody using. We sell tools for both interactive and automatic use. We are not trying to sell one over another. While automation is a beautiful thing in that it really shortens up design cycle time and minimizes resources, there are still some designs that can't be done with automatic routing. These include PDAs, cell phones and also PC mother boards, particularly for laptops where form factors are critical. Boards where autorouting has done a good job are monster sized boards, things like network switch cards, servers, mainframes, and big backplanes.
Would you expand on the types of products that have these complex form factors?
It's purely a function of visualization of the problem. Both automatic and manual tools are highly driven by constraints. The problem is that when you look at a board, designers can visualize where the best paths are for a parallel bus, where the components would go next to each other that most optimize. That's the sort of thing when you are in a very tight space. Human skill has an edge over the machine. The human can do things that the machine can not do as efficiently as possible. In these cases designers tend to choose the interactive approach as opposed to the automatic approach. I would point out that it is not a hard and fast rule, one way or the other. Most people are doing a combination of the two. Typically on a very dense design, they will lay it out or rather place it by hand so all of the components are in place. Then they will automate fan out and do a little bit of automated routing just to see how things are going to look. Then they come back and either manually clean it up or decide that they want to rip up a whole bunch of stuff and reroute it manually. It is kind of an iterative approach between automatic and manual tools. The reason we distinguish is because the automated layout side those boards are the ones where you tend to be able to push the button and get fairly close to 100% without a whole lot of manual cleanup.
Xtreme technology is the technology in play as it is with XtremePCB. The same patented technology is underneath both products. It allows simultaneous processing by multiple users on the same database. This is the unique thing within the PCB layout space that we have, namely the ability to automatically let multiple users work on the same database without having to go through a manual partition of the design and manually gluing back together at the end of the process. This solution can work across the network either locally or wide area, single continent or multiple continents. The server looks at what is happening with each of the users, receives their edits back, performs some rule checks to determine if they are still valid, makes modifications if necessary and then sends them back to each of the clients in real time. It is a very dynamic process to sit there and watch multiple people working concurrently and see the edits of each of their peers showing up next to them. That's the underlying technology.
We launched XtremePCB in late 2004 for this predominantly manual mode where multiple layout designers can work on a design and get it done quicker. In that case we saw cycle times being reduced by 40% to 70%, fairly significant considering the densities and the times taken to get some of these boards done. The types of things users would most likely to do by hand include component placement, the interactive routing aspect of it and other tasks like documentation, manufacturing prep. In some cases engineers are virtually looking over the shoulders of the layout designer, potentially even making edits to insure optimized performance on a circuit.
Over the past year how many systems of XtremePCB have been sold?
The number of systems sold is in the dozens at this point. These are the sorts of tools that fit into large enterprises but we have also seen service bureaus because it can cut down their design time. This is a key element of competition.
Some customers have experienced 40% to 70% optimization of performance. We are seeing better in some cases but we haven't characterized all the data we have been getting.
The new product is XtremeAR?
XtremeAR takes the same methodology that is used for XtremePCB and applies it to distributed autorouting, people working off of a central network with a central server involved. In this case it is a single user driving multiple machines. It is utilizing the same patented technology with a limit of 15 CPUs. We are experiencing up to this point roughly 10X reduction in execution time. This is huge compared to the incremental improvement that we can make in tools over the year and get improvement of the order of 10% to 15%. Here you have an immediate 10X reduction in cycle time just be leveraging existing hardware. This is with 15 CPUs.
The reason why it doesn't achieve 100% optimization is because of that patented technology where the autorouter looks at the design and partitions up these nets and sends to each of these processors. The routes get done by those processors, are received back, design rule checks are done, modifications are done if necessary and then redistributed to the autorouter. That kind of pulsing process is what reduces the optimal performance from 100%. It is not bad by any means but it is not network traffic or anything like that, although that can impact it. It is the router itself crunching away on circuits.
Where are people doing autorouting?
Largely digital, high speed constrained nets where they can do that sort of push button autorouter. The other place is where they typically want to do designs by hand. Here's the crossover in methodology where they may say on that motherboard for a laptop, I fully intend to route this thing by hand because of the density but what I would like to do is to figure out roughly how many layers do I think I am going to need for the circuit. Because it is a high volume application for the consumer market, cost is everything. Going from 2 layers to 4 layers to 6 layers impacts cost significantly. They can use XtremeAR to autoroute, automatically to evaluate overnight. They kick off a run with 2 layers, one with 4 layers and one with 6 layers and run them overnight. In the morning they come back and see that the 2 layer run got only to 50%. They would say that I can't route that any better manually, so I have to throw that out. The 4 layer run got to 80%. That's looking pretty good. The 6 layer run got to 100%. In fact the autorouter did it all by itself. I am betting that if I cut it back to 4 layers, I can save myself some cost and hopefully get it done. It comes back to virtually prototyping of the layout to get the most efficient design.