Growing System Complexity Drives More IP Reuse

IP reuse of both third-party and internal IP is growing, but it’s also becoming more complex to manage. There is more IP being used, and more systems into which it needs to be integrated, combined with other IP, and tracked throughout an organization.

In some cases, this is an economic requirement. In others, designs are so complex that engineering teams need to focus on where they will make the biggest impact with the highest value add and maximum differentiation.

“It is a widely accepted adage that chip companies try to innovate their way out of slumps,” said Steve Roddy, chief marketing officer at Quadric. “In practice, that has meant trying to refresh product portfolios while simultaneously tightening or holding the line on engineering budgets. If headcount can’t grow, the use of IP must grow to speed up the design process. That, in turn, leads companies to focus on their real sources of value-add and differentiation, not their hoped-for sources of differentiation.”

Processors are among the most complex to create, and the first to be used when staffing is tight. “In prior downturns we saw surges in the licensing of CPUs (in 2001, sales dropped 30% from the prior year) and the phase out of numerous proprietary CPU offerings that were legacies of the 1990s efforts by semiconductor companies to engineer internal me-too offerings,” Roddy said. “Similarly, the 2008 to 2009 decline (12%) saw surges in licensing of commercial DSPs and corresponding internal DSP architectures de-emphasized. The anticipated 2023 market plunge after the pandemic-juiced peak years of 2021/2022 is likely to lead to a widespread re-evaluation by chip companies of investments in proprietary machine-learning accelerators when there are numerous licensable NPUs and GPNPUs available from the merchant IP vendor community.”

Business decisions drive much of the activity surrounding IP. “We are definitely seeing that re-using IP, and the desire to re-use IP but not knowing where to find it, are getting a lot more interest of late,” said Simon Rance, vice president of marketing at Cliosoft. “According to the big companies we’re talking with, one of the reasons is multiple acquisitions. IP exists in one place of the organization for whatever company they acquired, and in a different place for everybody else. What they’re trying to do is minimize costs and improve time to market — all the usual goals. They’re trying to find more IP to re-use, but the first stumbling block is they don’t know where it is. Other people in the company don’t know where it is, or even what IP actually exists, and that has to be solved before they can even contemplate re-using it.”

Björn Zeugmann, group manager integrated sensor electronics at Fraunhofer IIS’ Engineering of Adaptive Systems Division, noted that IP reuse has become increasingly important in recent years, and that more complex designs can be completed in the same design time only with increased reuse. Design time would explode if every IP or basic building block has to be created  from scratch for every design.

Reuse also adds reliability, because that IP already has been vetted in silicon. “Once verified and measured, IP also has to work in another more complex design, which raises its overall design reliability,” Zeugmann said. “What makes reuse difficult is handling a wide spread of technologies and flavors. Porting of designs and new verification is daily business, and it’s more efficient than building from scratch. But ported designs often cannot push the boundaries of new technologies, so this methodology is suitable primarily for basic blocks. High-end designs always need more modification, or to start from scratch. It’s challenging to handle the data management over a lot of IPs in different technologies and libraries. Here, a common internal database combined with knowledge of external ones is the key for efficiency.”

Build vs. buy
Pre-developed and tested IP is almost always a faster way to market, particularly for standard IP where building your own interface or I/O adds little differentiation.

“It used to be that all of our value messages were around why you would buy from an IP vendor rather than building it yourself,” said Mick Posner, product line senior group director for high performance computing IP at Synopsys. “That still exists, but it’s for a smaller percentage of engagements we have now because IP is so widely accepted. It delivered on its value of accelerating deployment. The uptake comes from the evolution of those standard protocols. If we just look at interface IP, which used to be on a four-year protocol spin, you would see that cadence. Now, that has compressed. Protocols now average two years, so the value of IP becomes even higher because no one developing an SoC is going to be able to rev their own stuff that quickly. Reuse is the only way to keep pace with that technology.”

That becomes more apparent as design complexity increases. “You’re not re-using just a standard cell or D-to-A converter,” said Sushil Sharma, director for IP technical solutions at Synopsys. “It’s much bigger blocks — either hard IP, or a much bigger soft IP block. So as the complexity of the IPs grows, the design cycle time is still the same 18 months or 2 years. Yet now they have to deal with very complex IPs. These may be in-house IPs, or they may be IPs licensed from vendors like Synopsys, but they are very complex and there are many of them. Part of the IP management problem or challenge customers are looking into is what else they can get from the supplier of the IP internal or external to help them manage that complexity.”

So rather than just developing reusable IP, one of the goals is to make it easy to implement in one or more designs. “As an IP supplier, we do everything we can to make our IPs reusable, including a lot of programmability,” said Ashraf Takla, CEO of Mixel. “Our methodology is built on a modular, ‘LEGO-rithmic’ approach. Everything is built LEGO-like, meaning schematic layout, testability, architecture. That’s what we do to make our blocks reusable. Then, we build the IP itself to maximize the ability for many customers to apply it to their designs.”

There’s always a balance something that is silicon-proven and serving a wider market. “Maybe they used it previously and want to re-use it,” Takla said, “so while we do make every effort to make our IP reusable, customers also would like some differentiation. The foundation is there with added customization to that IP.”

The critical piece to getting this mix right is listening intently to customers. “We take their feedback very seriously and act on it. In many cases, it leads to a better performance IP for Mixel. We have a patent on one of the MIPI implementations that basically was based on a requirements from one of our first automotive customers to implement some testability. That’s the cycle that happens.”

Locating IP assets
Reuse often is considered to leverage FTEs, said Benjamin Prautsch, group manager mixed-signal automation at Fraunhofer IIS EAS, adding that IP management is crucial for a good overview.

“IP management systems help to organize IP versions and design data,” Prautsch said. “However, organizing relevant meta information across projects is tricky. Holistic IP management that also includes, for example, specification tracking, customer interaction, technology management, or measurement results, is still a topic for discussion. A variety of tools are used, but a well-organized ‘dashboard’ approach is missing. Thus, IP is often hidden in a ‘designer’s shelf.’”

That requires a system to find the IP within an organization, said Synopsys’ Sharma. “If I’m a design manager at a large company, and I need certain IP, what do I do to find out whether my company already has this IP? The process varies from company to company. Many of our large customers typically have internal teams that are responsible for this. Some are very well organized, with very well staffed teams. They keep track of all the IPs they license from various vendors, including versions and other information. In that case, the design managers go to that team and say, ‘I need these IPs. What do we have?’ They then give them all the details.”

But the picture is very different for new startups with fewer than 20 engineers. Sharma said an IP reuse or IP management team is important to establish. “What do we see customers doing if they are in that category? Part of the answer is what they would expect their IP vendor to also offer. For instance, if I’m a customer of Synopsys, I can log into a customer support portal and find out what entitlements are available for which IPs. The same portal provides the latest version, details on the process technology it is part of, as well as release notes and the PDK version that’s applicable.”

In some cases, it makes sense to leverage a commercial IP management tool. These tools allow the design manager to point to the design spec for the next system and the types of IP a company has.

“Basically, it’s almost like going on Amazon,” said Cliosoft’s Rance. “You can put in a search for a specific type of IP, and the IP management system will tell you if there is any IP like that in their company, and will list them out. It allows the design team to compare IP side-by-side just by data, which can be by power, by performance metrics, and by how many times that IP has been used in previous designs throughout the company, as well as to give insights on the level of quality. It also can show if the IP has been used in real designs, and so on. It allows the team to make the better decisions quicker, so they can say, ‘Narrow it down to these two IPs. Now let’s go test them out, play with them, see how they look in simulation and so on, as well as do some real tests around it.’”

IP is typically categorized in three types — third party, internally sourced, and consumer IP.

“Third party IP includes IP from Synopsys or Arm they have active licenses for,” he explained. “Internal IP is also referred to as corporate IP, which can be re-used if it can be found. But that IP cannot be modified because it has been fully verified to corporate level standards, and that particular piece of IP meets the low power, high performance, or other requirements of, say, Broadcom for their Bluetooth. Consumer IP exists within the company. It may have been used on previous designs. If they can find it, designers can take it and modify it to meet whatever the new system requirements are.”

Rance noted that a key reason why IP reuse has failed in the past is that it’s almost at the register transfer level, or schematic level if it’s an analog IP. “With what we manage, we take it all in, so it’s got the regression tests, it’s got the design specs, it’s got all the PPA-type data and results for that IP. We’re even starting to see some of our customers using this for chips that have taped out and they’ve used our product for. They’re putting in data from the field that can be analyzed at some point in the future to improve a future release of that, as well. That’s the other aspect of what you get from that search result that’s not just, ‘There’s a piece of IP that exists.’ It even includes design engineers and architects chat sessions, or documents that say, ‘Hey, this one didn’t really integrate very well with other IP in the system, so we had to add this glue logic to make it connect and integrate at the system level.’ And that glue logic is also stored there with it in case somebody else in the future wants to just take it as is, or tweak it, and they don’t need to reinvent the wheel trying to get it to integrate.”

Licensing still an issue
Any IP management system must have the ability to keep track of legal usage license. Synopsys’ Posner said that in the IP space, business models and licensing have not changed over the past  20 years. “Predominantly we sell IP as per project. That fundamentally means we are protected by legal agreement. I cannot tell you how many times we find out that IP that was licensed for Project A is now being used on Project B. Maybe it was a full rev version B of the product, or maybe a completely different group sees it there and re-uses it, even though they’re not fundamentally legally licensed to use it. That is still a big concern.

Making sure everyone stays updated within an organization regarding the IP or subsystems they have access to is essential. It’s also one of the reasons why IP re-use has not reached its business potential.

“What is necessary is to implement rules and checklists, where if you design a piece of IP from scratch, before you release it as a piece of IP into the company or for a system design you must go through these checks. Does it have the accompanying spec and documentation? Does it have the verification, scripts and test environment and setup and the like? That exists with the IP. Since every user organization is slightly different in their requirements, workflows can be created to ensure all of this. One example of that would be I if I’m releasing a piece of IP, I can say, ‘I’ve met all of these checklists that have been put in place by our company for producing a piece of IP and releasing it.’ When I say it’s ready to be released, a workflow will kick off that will then go to my management hierarchy, where they’ll get emails and so on to say, ‘Please review this. Make sure it’s met all of the requirements and checklists and legal requirements.’ Then it will go to legal, for example, as the next part of the workflow to make sure this piece of IP is classified correctly for export control. That helps overcome the common problem of a designer creating a piece of IP and saying they are too busy to really put it all together for somebody else to re-use in the future because they’re moving on to something else, another IP, or another design. It safeguards that to make sure what they’ve done has the opportunity of being found or re-used by the company in the future.”

Further, in the overall scheme of IP reuse and IP management, it means more of the burden is on the IP supplier to fill in the gaps, which is another aspect of the complexity of reuse to begin with.

“Let’s say I need to have a display block in my SoC,” Synopsys’ Sharma said. “That display block is not usually available as a single piece of IP. Usually you will have a PHY and a  controller. You may have a few other things — some bridges, switch fabric, etc. Who is going to put all that together? If I am a project engineer, what do I do? That’s where I would say, generally, we see many customers are not only looking for very well proven IP for the new interface protocols that are evolving more quickly, so IPs are evolving, but also can the IP supplier offer a next higher level of reuse where these things are put together into customer-specific or design-specific subsystems? This would ease the challenge of putting them together not as a generic IP block. If it was a generic IP block, an IP vendor could simply offer it as an IP. But something that’s put together in a context-sensitive, application-specific or configuration-specific way and given to the project team is not just about IP blocks. It’s about subsystems, which are created in a customer-specific way. The next question after that who is going to harden it?”

This points to the general trend of IP users looking for a complete solution. “That means it’s not enough to just provide RTL,” Sharma said. “It’s not enough to just provide GDS, or even a test bench and a VIP. Design teams are now looking for additional things, such as whether we can provide IBIS models or help to do the signal integrity analysis at the chip level. Or, ‘Do you offer something to assist with prototyping, because I need to prototype this in a Zebu or an FPGA on HAPS?’ These additional aspects are coming up as part of the overall effort to help design teams manage the complexity, effort, and additional challenges that come up with this increased amount of IP to use, and the increased complexity and layers of abstraction of IP reuse.”