Looking at the logo I can see something that looks like a bridge.
How many employees does Ponte have?
About 70 now! Probably 10 to 15 are in the US. The bulk of them are in engineering in Armenia. It turns out to be a fantastic place to have a R&D team because first of all it has a cost structure that is less expensive than the US and second of all it is a very talented team. There are a whole lot of mathematician and software developers that have come out of Armenia and the former Soviet Union. I would characterize the people that I have met from Armenia who work for Ponte as extremely hard working, dedicated, motivated every bit as much as anyone I have met form India or China. It is a pretty good situation there.
Would you comment on the number of customers and revenue that Ponte has?
I don't think I can announce what those numbers are. We do have revenue. We are shipping product that customers pay for. We are in a production mode. We have a set of customers using the product, the first versions of releases. We haven't officially announced the product, we have announced the company but the product announcement will come up in the next couple of months. At that time we will give some details about how people are using it and what impact it has had.
Your web site does not have a description of the product.
I will tell you in essence where we're headed which is developing a yield model. Usually the way yield has been handled on the design side is through design rules. The issue there is that the fab has simply said that if you follow these rules, the design will yield. We all know that this is breaking down. Yields are not high at 90 nm and below. The solution to that is to provide a model instead of rules that provides the tradeoffs. If you can afford to do this, maybe spreading things out a little bit more, you should be able to get more yield, maybe doubling up on vias. There is a whole set of techniques there. We are trying to capture that in a model that will cover a wide variety of different kinds of defects. That way with our first tools and analysis design teams can look at the design and make some quantitative judgments the yieldability of their design work.
The rule based approach to DFM is dependent upon process node, the fab and so forth. Are you models also dependent on these factors or are their generic?
The models are generic because it is physics. But to make them really work, you need foundry data. One of the things we haven't announced the particulars about is that we have cracked the foundry model. There will be foundry data available for use with our tools. Our customers have told us that this is a requirement. They have helped us working with the foundries to make that data available. That's actually one of the big barriers that you hear when you talk to small companies that are in this space. You are going to find that having the foundry data is a critical element to the believability of the analysis. In general the foundries don't want to give out their defect data because they will essentially admit that they have failures. Everyone knows that they have failures but they don't want to publish that data. We have come at them from the standpoint that the customers want it because they need to be able to predict what their costs are going to be and we've also built a model around it where there is encryption. So the actual data is protected.
Are there any competitors coming at the problem in a similar way?
Not that we are aware of. There are a lot of companies that are focused on what I would characterize as second generation. I am not aware of any companies focusing where we are on being the design side, building this statistical yield model.
What is your sales model, direct or indirect?
A combination! We have already signed up two distributors. We will probably make some public announcements about that but it is not a secret. A group in Taiwan called Markettech (MIS) and a group in Korea called Win Technology. We use a combination of direct and distributors.
What is the packing of the product (development license, royalty, )?
At this point we are not going to reinvent the wheel. We are going to go with a straightforward EDA software license, primarily subscription. A small portion of customers have asked us for perpetual license which we will have as well. As a small company we are not going to attempt to change the market to some different formulation.
What is the price range for the product?
It will be in the $100K range to get started. The product announcement will have more specifics in a couple of months.
Would every design engineer have a seat or only certain experts?
The way I characterize it, it is really at the physical design level not at the RTL level, not at the ESL level. It could be used as early as someone is doing floorplanning. It can be used in library development and used at the system design level to make tradeoffs. These are the three areas that are most likely to use it.
How does DFY relate to DFM? Subset? Complement?
We touched on that earlier in terms of the three generation and how the third generation is actually DFY. My belief is that DFM is one of those terms that has become somewhat meaningless because what does it mean to design in order to be able to manufacture. If you are not designing it to be manufactured, what is the point of designing it. The issue is more one of cost and bridging from the manufacturing data into the design side. The first two generations are purely on the manufacturing side. A designer doesn't need to know what the mask actually looks like that is going to print his circuit. A designer doesn't really need to know where the process windows are on exposure, aperture and so forth. They just assume that if I follow a set of rules, I should get yield out the back end. DFY is that third generation where a designer can get data in a model in a manner that is meaningful to make tradeoffs that impact yield. Manufacturability is not a binary where it works or it doesn't. It is a statistical event that happens.
Armed with this data what steps can a designer take to increase the yield of his design?
It's a little beyond our tools. Out tools are analysis tools. For instance, we can look at different floorplans of a chip that are all possible at trial layout and then run an analysis to see which one more likely to have less yield loss, i.e. will yield better. You can also look at library elements, different figures of merit on different ways of laying out a cell and which one would be yielding well. One of our early customers looked at the library elements of a very large chip. They determined that something like a dozen library elements were each used 50,000 times. There were some yield issues within those elements. By going back and resigning those elements without a penalty in area they were able to increaser the yieldability of that overall chip. The analysis is to find the issue, figure out something to do with it and give yourself a benchmark of whether you can improve or not. It could be at the floorplanning level, the aspect ratio of the chip, or it could be down to a cell level. It could be a library element.