中文 English

Making IP Friendlier

Experts at the Table, part 2: Some IP issues remain unsolved even after 20 years. What the IP industry needs to change for growth.

popularity

Semiconductor Engineering sat down to discuss IP tracking and management with Ranjit Adhikary, vice president of marketing for ClioSoft; Jim Bruister, director digital systems (since retired) at Silvaco; Marc Greenberg, product marketing group director at Cadence; and Kelvin Low, vice president of marketing at Arm. What follows are excerpts from that conversation. Part one can be found here.

SE: A lot of people talk about difficulties getting IP in the Cloud, but if the Cloud is nothing more than an extension of someone’s data center, how does IP even know it is in the Cloud?

Bruister: There are two aspects to this. One is that it could be a database or repository. The second is that people are actually starting to do collaborative design on the Cloud in which case, a design is a collection of IP, so the person doing design on the Cloud will say I want this IP from here and this one from here and then will assemble it all in the Cloud and simulate it.

Greenberg: We can create virtual chambers as part of the Cloud for a customer. IP can go into that, tools can go into that, they can do their design. One interesting aspect is that we have a requirement, sometimes for defense projects or automotive projects, where designs need to be stored for a period of time. We can take a whole virtual machine and store that for future use, and that could be deployed on a different Cloud platform. There is a little bit of a paradigm shift that people have to get used to. The server is not on my desktop or in my server room. It happens to be somewhere else. There is security because there are licensed products going in and coming out. Once you accept that, designing in the Cloud should be no different from designing with tools in the server room. You just have more managed tools, managed tools versions, CAD people in the Cloud who are taking care of those things for you.

SE: Where are the pain points in conducting IP transfer?

Greenberg: Doing it securely. I want to create a secure connection between me and wherever my customer chooses to receive their IP. I don’t want it to be intercepted by a third party along the way and I don’t want a third party to be able to inject something that should not be there. So, you need a secure connection between the end points.

Bruister: It’s the business model and licensing. Every license agreement is different, done by a different set of lawyers, with no standard license agreement.

SE: There were attempts to fix that over two decades ago.

Greenberg: Yes, the virtual component exchange – VCX. We based that on top of standard contracts that were used in Britain for all kinds of electrical contracting.

Bruister: That could have been the problem.

Greenberg: It was a good thing. The Scottish government was willing to change the local laws to better suit the needs of the IP industry. That is one of the reasons why they located it there. They tried to base the standard IP agreements on standard electrical contracting agreements. There have been standard agreements for electrical contracting for more than 100 years. If you are going to install electrical wiring in your house or a lamp post, there were standard contracts in place for it. We tried to do the same for IP and as an industry. We just weren’t ready to do it.

SE: So how come, 20 years later, we are still not there?

Greenberg: Because there are corporate lawyers in every company, and lawyers never see eye-to-eye.

SE: The automotive industry is putting more demands on IP, such as certification. Will that create tiers of IP such that you can buy the same IP with or without certification.

Bruister: I don’t think it is that clear-cut, but it may be true in a very general sense. Most IP vendors came from either small design groups or in-house teams. Some have worked on certification, some have not. If a customer wants them to create a certified block, they will work on it. People do not tend to design anything before they have a customer for it.

Low: In the automotive area, there are tiers of certification requirement. The smaller the number, the more challenging it is. There is a lot of additional documentation required. Companies are learning the needs of the industry. At the same time, the automotive industry and their expectations are changing. OEMs are looking to the future and their requirements are different. From IP design, reliability considerations versus consumer products are quite different. They are two sets of IP, and there could be different tiering depending upon their grades and products could be marketed different. We do have to charge a premium because of the additional investments.

Greenberg: We try to take a structured approach to it. We had a full set of IP for 16nm last year. We have select IPs on other nodes. We try to make as much of it standard as possible so that for each IP there are configurations where we have taken it through the process. Some customers will want something different. So they can take what we have already done, or if they require a different set of features, we can do it. But it is a process of having to re-certify. Another difference between commercial and automotive IP is that automotive requires additional safety features that do not exist in the commercial IP. There is no necessity for them on the commercial side. But for automotive, they are necessary to meet the ASIL requirements.

Adhikary: From an IP management perspective, when you classify an IP there are certain attributes that you want. When new things come along, like automotive and different certifications, people want to be able to find IPs easily, and certifications keep changing. There are different certifications for different IPs. This makes it difficult if you have a hard-coded method of characterizing IPs. Changing it can be quite messy. So there is a complex matrix of information, such as different process nodes, certification, if patents apply, etc. With the amount of information out there, it becomes challenging to make sure everything works and is easy to use. When you find multiple IPs, how do you compare them? You need to know all of the details in order to be able to make the choice.

SE: There is the IP provider and the IP user. There is nothing in between. Why do we not see someone take a number of IPs and create a subsystem out of them and sell that?

Bruister: It is very segmented between the provider and the user. There are not many IP aggregators. From licensing, which is not a one-stop shop, the user would still have to go to every different IP company. There are some people who want to do that.

Greenberg: ASIC vendors will sometimes do that.

Adhikary: The problem is that for big companies, they get a better rate for the IP directly from the IP provider compared to a small company. Even if a small IP company is providing an aggregation service and provides a complete IP sub-system, the price may not be right. These are issues that come into concern.

SE: If I take an IP block from each of you and integrate them, I have put value into that, and you don’t even know what price I am paying from each block.

Low: The ASIC provider is the closest to that. We are seeing a trend, and ASIC is coming back. Some system companies want to create more optimized solutions. Over time, some of these companies will build in-house silicon capabilities. But for now, the ASIC providers are servicing them.

Bruister: A couple of years ago I did a presentation titled, ‘Why is chip design so hard?’ The biggest reason is there needs to be a general contractor for doing chip design. Just like building a house, you don’t go buy a toilet or a sink or all of the pieces. You get a contractor to do all of that for you. We need that for the chip industry. ASIC designers are kind of there. If you look at ASIC houses, if they have a customer who wants a chip and says, ‘I need this IP and that IP,’ he will go and do all of the leg work. But in the background, he is having to go through all of these contracts for you. So it is just pushing the problem down a level.

SE: Chip design never got to the point of having nice encapsulations, where you could easily put blocks together. You know you may be losing 5% to 10% efficiency or area, but that has never been acceptable to the hardware community, whereas software teams are quite happy wasting lots of energy or performance.

Bruister: I don’t think chip designers should be doing that. You should not be looking for everyone to be a chip designer. You need to enable people to do designs who are not chip designers. In the old days, they were doing layout and transistor level design, but when you got synthesis tools, people weren’t drawing Karnaugh maps anymore.

Greenberg: You can throw everything at a giant CPU but ultimately, the CPU does not do a good job for everything and then you need some custom accelerator and you get back into the IP domain.

Bruister: In the IoT space, where they are talking about billions of devices, there are not enough designers in the world to design them all. How do you enable different people to do that?

Greenberg: There is a lot of commonality in IoT applications that could be served by a relatively small number of chips. People still go off and design their own chips because they are looking for that last few microwatts of power reduction. If there is a piece of dead silicon, they want to get rid of it because that costs money in high volume. We always see this process. We see people start prototypes using FPGAs, and while it doesn’t run as fast as custom silicon, will be more expensive but that is fine for the first limited production run. If successful, they will go to ASIC or even COT and constantly try to reduce cost and improve performance. But with FPGA, you do not need to know all about the back-end stuff.

SE: What’s the one thing that you want to change.

Bruister: I said my piece. Legal as well.

Greenberg: We spend a lot of time with lawyers.

Adhikary: The biggest problem we see is getting engineers to provide the required information. Everyone believes they do not have the time, but if you do not take the time to ensure that for any design, or change that you made, that information becomes noise to some extent. I would like to see a mindset change in engineers. So many people are caught up in reactive things. They just want to get this chip out, rather than saying from a holistic standpoint, ‘What can I do to make things more efficient?’

Low: It is always easier said than done to get companies to collaborate. We talked about abstracting the solution. It is not new. It is a reference design. We show how each of the IP works, and the end customer can feel comfort and confident. Business models are harder to align across companies. Having companies get together more regularly helps, but it takes a lot of time and energy.



2 comments

Kevin Cameron says:

20+ years is about how long since Verilog-AMS came into existence, it should have been a full spectrum modeling language that would let you ship out behavioral models of IP that would work everywhere.

Cadence hobbled that language, and the digital guys refused to integrate it properly.

The guys above have had ample opportunity to fix it or do something better, but they just sit around and gripe about how the ecosystem doesn’t work.

Note: happy to take their salaries and actually fix the problems.

Roy Stahl says:

Good articles, Brian. You pulled together a great panel openly willing to discuss the problems with IP management in the industry. It is something I have spent time wrestling with over the last few years as libraries keep getting larger, and legal agreements keep getting more complex. With more IP qualifying for export control (3E002A and 5E002), it is going to have to be dealt with in a more serious, strategic manner.

I think contracts will only get standardized when margins get squeezed to the point where companies can’t afford NOT to (lawyers aren’t cheap). Designing in the cloud is the raging storm. Designs in the cloud have to take advantage of data sharing of packages like libraries, but security has to be locked down extremely tight.

Finally companies will have to track projects and IP more assiduously at a high level where legal license restrictions, version control, and people assigned to projects can all be viewed (audited) in one place. The industry doesn’t have another 20 years to make this happen.

Leave a Reply


(Note: This name will be displayed publicly)