Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/should-eda-follow-a-foundry-model.10435/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Should EDA Follow a Foundry Model?

A

Arthur Hanson

Well-known member
Just like a semi foundry takes knowledge in executing making chips for a variety of customers and shares it, yet keeps each customers information separate and private, should not an EDA firm be set up in its own cloud to share the expertise that they develop from monitoring a large number of separate process for different be used to improve the processes for all their customers? TSM has done an excellent job of keeping individual customers IP separate and private, but uses the improvement in process information to the benefit of all. Would not this process if applied to EDA speed up the evolution of the design process to the benefit of all through the use of big data. If TSM can keep proprietary information separate and confidential while spreading process improvements, couldn't EDA firms use the same structure to benefit their customers as well. Auditing the process on a real time basis could assure security while giving the customer the best practices on a real time basis. This could also be done on a virtual machine bases with most of the process done at the customers site, although this would be unwieldy and cumbersome compared to a private cloud. Any thoughts, comments or observations on this appreciated and solicited.
 
Arthur,

TSMC manufactures physical devices, chips, packages and provides testing services. EDA is obviously a software business model, not really comparable to selling silicon like TSMC does. EDA companies try quite hard to get their users to share information in order to improve the tools, however even supplying a test case to reproduce a software bug is an issue because IC designers don't want any piece of their IP to leave the facility. For DFM and DFT tools, yes, there is a symbiotic relationship between IC designer, EDA vendor and foundry to help pinpoint IC layout areas that are susceptible to yield issues. Some of the largest EDA customers will have AEs on site at the customer to help reproduce the bugs in the software while not letting the actual IP out of the facility. In general though, the IC designers don't want the EDA vendors to know too much about what they are designing, how they are designing, or even how they are actually using the EDA tools because of concerns about privacy and keeping a competitive advantage over others.
 
Arthur Hanson said:
Just like a semi foundry takes knowledge in executing making chips for a variety of customers and shares it, yet keeps each customers information separate and private, should not an EDA firm be set up in its own cloud to share the expertise that they develop from monitoring a large number of separate process for different be used to improve the processes for all their customers? TSM has done an excellent job of keeping individual customers IP separate and private, but uses the improvement in process information to the benefit of all. Would not this process if applied to EDA speed up the evolution of the design process to the benefit of all through the use of big data. If TSM can keep proprietary information separate and confidential while spreading process improvements, couldn't EDA firms use the same structure to benefit their customers as well. Auditing the process on a real time basis could assure security while giving the customer the best practices on a real time basis. This could also be done on a virtual machine bases with most of the process done at the customers site, although this would be unwieldy and cumbersome compared to a private cloud. Any thoughts, comments or observations on this appreciated and solicited.
Click to expand...

The semiconductor business is very competitive and secrecy is a big part of that. Sit in a vendor meeting with Apple and you will see it first hand. It's like a classified briefing with redacted documents etc... a need to know basis and as it turns out EDA vendors do not need to know much at all. EDA vendors just need to work closely with the foundries and customers are good to go. TSMC is the security buffer and they are quite good at it. TSMC is the "trusted" foundry, right?

Apple would definitely be better suited if they were more open about their EDA tool and IP usage but at what cost on the competitive side? A double edged sword for sure.

EDA in the cloud is another example. Cloud EDA would save companies billions of dollars in IT cost and tool licensing efficiency yet here we are, using the inhouse EDA tool usage model of the 1990s. Again, security is the prevailing reason we are not cloud based.
 
Daniel Nenni said:
The semiconductor business is very competitive and secrecy is a big part of that. Sit in a vendor meeting with Apple and you will see it first hand. It's like a classified briefing with redacted documents etc... a need to know basis and as it turns out EDA vendors do not need to know much at all. EDA vendors just need to work closely with the foundries and customers are good to go. TSMC is the security buffer and they are quite good at it. TSMC is the "trusted" foundry, right?

Apple would definitely be better suited if they were more open about their EDA tool and IP usage but at what cost on the competitive side? A double edged sword for sure.

EDA in the cloud is another example. Cloud EDA would save companies billions of dollars in IT cost and tool licensing efficiency yet here we are, using the inhouse EDA tool usage model of the 1990s. Again, security is the prevailing reason we are not cloud based.
Click to expand...

Maybe it's time for a change. TSM can maintain a very trusted level of secrecy to the point everyone uses them, why not an EDA firm. I'm sure TSM has a security structure that isn't unique and security structures have been set up for sensitive IP and just plain secrets in a way information is compartmentalized. It's just the difference between a physical construct and a virtual one. I know Samsung abused this with Apple, but they were are bad actor in this case. Is there not the talent out there to design a comparable compartmentalized system where just the formulas and not the actual proprietary IP is visible. Throughout history many systems that handle sensitive IP have been designed so there can be a degree of collaboration without revealing key information. It would be like living software and with the right system could advance tremendously, benefiting everyone. Customers would soon learn who to respect and not respect, just like many other businesses. An ideal everyday example is a lawyer who perfects his techniques working for several clients in the same area, but not sharing their cases, but learning IP that benefits all the clients. It could even be done by a black box system so no one would know what was actually done, but improve the process by a set of formulas. Any system will require some degree of trust otherwise the foundry system would not exist. Is there not someone with the standing and respect of Morris Chang in the EDA industry? If not, that's a very sad commentary on the industry and the people in it. No system is perfect, but I have no doubt one could be built within acceptable tolerances. IP is IP whether physical or virtual.

There are systems out there that handle life and death on a massive scale information and they have been able to maintain security, it should be possible in the case of EDA. I have personally been exposed to such structures that handled information far more critical than EDA, so I know it is possible and they can be built.
 
Last edited:
Arthur,
To some degree that is happening today, though not always in the context of the EDA vendors. Chip builders that are content with near-optimal results using standard IP at a specific node can avail themselves of pre-built and tested flows for their combination of silicon IP / node. Or they can work with fabless ASIC companies who have evolved tuned flows for specific end-architectures. The two challenges I see:
* Most aggressive design customers view their verification and implementation flow plus some of their specialized IP as their largest value-adds. They do things differently to achieve differentiated end product.
* There have been many attempts at EDA in the cloud / EDA as a service but financial and IP protection forces always deter. There has to be sufficient new and relatively immediate benefit for one or more of the parties (EDA vendors, IP vendors, foundries, customer) involved to pull it all together.

I'll also point out that EDA vendors, IP vendors, and foundries have evolved a level of engagement together and with end customers to give the customer the ability to do even leading edge designs in the comfort and privacy of their "own home".
 
Last edited:
The power of a platform over a stand alone system has been proven over and over again. Two that are known for security and privacy are Apple and TSMC. Apple is expanding into medical where confidentiality is not only demanded by customers, but mandated by law. There are currently international interagency intelligence platforms that share and handle information with the utmost security and secrecy. There are also large criminal empires that do the same thing. I want the good platforms to win. Platforms can also be abused as has been proven in the case of Samsung and Facebook when they don't take privacy and compartmentalization seriously. A good platform can benefit everyone as TSMC has done to advance literally the technology of the whole world.
 
Great discussion, thank you guys.

Just a little history, the first wave of commercial EDA companies spun out of semiconductor IDMs who were trying to cut costs. This and the fabless business model pioneered by TSMC created the massive fabless semiconductor ecosystem we have today. A similar thing happened in IP a few years later.

These first waves of EDA companies brought both tool and design expertise to customers and took a leadership role through tape-out. Today EDA companies are in a supporting role since customers now have much more design experience and again secrecy is a competitive necessity. Additionally, as fabless system companies (Apple, Amazon, Google, etc...) come into play they want to have even more control and secrecy so EDA and IP companies are definitely in a detached supporting role.

Today the foundries act as a buffer between customers and EDA and IP companies making sure the tools and IP work well with the process. TSMC for example uses customer test chips for process development so they know what is coming and can better prepare the EDA and IP companies without spilling the secret sauce.

One of the possible EDA disruptions would be if Intel or Samsung bought an EDA company and made those tools "free" for customers by taking an EDA royalty fee on the wafers. They could also put the tools in a private cloud. Do you remember when Artisan introduced a free royalty based model backed by the foundries? Disruption is for the greater good, absolutely!
 
Arthur,
This is partially a "chicken and egg" problem. Today there's not enough economic benefit in the platform to overcome some of the legal and business forces that prevail at successful companies. More specifically, most big or innovative companies (EDA, IP, foundries, end-customers) require a range of near-unlimited legal liability and indemnification from parties taking on their most confidential and proprietary information / IP. In spite of excellent security and sharing solutions available today, these companies are wont to back off on these liability and indemnification requirements. Conversely, companies interested in bringing that protected IP in-house are only willing to take on the legal liability if there is an overwhelming economic win for them (i.e. an Apple deal). Your premised EDA cloud platform is dependent on an EDA vendor being willing to take on that kind of liability from potentially several leading edge foundries, several IP suppliers and end customers. Where's the deal that would make that a big enough economic win for the EDA supplier ?

Having seen many of these NDA and IP licensing agreements over time, I can assure you that most of the major sticking points were things that required CEO/COO level approval to satisfy the legal team it was worth proceeding. Typical examples:
- unlimited liability / indemnification
- unbounded risks - ambiguity in what might trigger liability or breach. Typically happens when the IP owning companies are unwilling to translate legalese into more specifics. For example, corporate lawyers prefer to keep breach criteria in legal terms that would have to be sorted out in court, rather than specifying allowed and disallowed uses up front.
- unenforceable operational restrictions - things like "anybody who touches this IP can't work in other parts of the company for 5 years". Even if that is operationally possible, it may not even be legal from an HR perspective.
- cost prohibitive enforcement - In some agreements, the IP entitlements can get incredibly complex - this needed list of people can do thing A with the IP, this list can do thing B, other groups in the company can do thing C with just one part of the IP, but this other group can't even be aware that this IP exists (true example).

As IP from all parties age and move from leading edge to mainstream, the legal requirements become less onerous. But the monetary motivations also drop in the mainstream since the end-customer becomes more price sensitive.
 
Last edited:
I think this is what you are seeing with MIPS-V which is not just the cores but a growing ecosystem of open source IP. And you do not even need it to be open source for the model of a common EDA and quality control/refinement loop to make sense, there are IP blocks widely used like SERDES where the whole community could benefit and the IP owners could license only on condition of getting certain kinds of feedback into their quality control and future designs.

The ecosystems are more powerful than secrecy, in the long run. A web of shared infrastructure hat even frenemies can share is based on mutual benefit. The typical ASIC or SOC is 20% secret sauce, 80% shared ecosystem. At the process level this has been a strength for TSMC. At the system level, it is why Open Compute has multiple cloud companies supporting it. Those who figure out how to share will in the long run have an advantage. You can see how this has worked by looking at open source software. There are many subtle lessons there about how it really works, and the license was just the starting point.
 
Tanj said:
I think this is what you are seeing with MIPS-V which is not just the cores but a growing ecosystem of open source IP. And you do not even need it to be open source for the model of a common EDA and quality control/refinement loop to make sense, there are IP blocks widely used like SERDES where the whole community could benefit and the IP owners could license only on condition of getting certain kinds of feedback into their quality control and future designs.

The ecosystems are more powerful than secrecy, in the long run. A web of shared infrastructure hat even frenemies can share is based on mutual benefit. The typical ASIC or SOC is 20% secret sauce, 80% shared ecosystem. At the process level this has been a strength for TSMC. At the system level, it is why Open Compute has multiple cloud companies supporting it. Those who figure out how to share will in the long run have an advantage. You can see how this has worked by looking at open source software. There are many subtle lessons there about how it really works, and the license was just the starting point.
Click to expand...

80/20% secret sauce is probably right, except when it comes to Apple. They license the ARM architecture versus using cores, do quite a bit of their own IP, and get a custom process from TSMC, I would say it is 80% SoC secret sauce and 20% shared ecosystem.
 
Daniel Nenni said:
Great discussion, thank you guys.

Just a little history, the first wave of commercial EDA companies spun out of semiconductor IDMs who were trying to cut costs. This and the fabless business model pioneered by TSMC created the massive fabless semiconductor ecosystem we have today. A similar thing happened in IP a few years later.

These first waves of EDA companies brought both tool and design expertise to customers and took a leadership role through tape-out. Today EDA companies are in a supporting role since customers now have much more design experience and again secrecy is a competitive necessity. Additionally, as fabless system companies (Apple, Amazon, Google, etc...) come into play they want to have even more control and secrecy so EDA and IP companies are definitely in a detached supporting role.

Today the foundries act as a buffer between customers and EDA and IP companies making sure the tools and IP work well with the process. TSMC for example uses customer test chips for process development so they know what is coming and can better prepare the EDA and IP companies without spilling the secret sauce.

One of the possible EDA disruptions would be if Intel or Samsung bought an EDA company and made those tools "free" for customers by taking an EDA royalty fee on the wafers. They could also put the tools in a private cloud. Do you remember when Artisan introduced a free royalty based model backed by the foundries? Disruption is for the greater good, absolutely!
Click to expand...

I think it would be interesting if the foundries, ie TSMC, got into the EDA game and charged a wafer royalty on it as you said. Better yet, a cloud based EDA tool that could also be used for ordering after designs are validated. If it could be integrated in a sort of design to manufacturing workflow, that would be amazing. Especially for smaller customers who are designing IoT chips and are focused on time to market, something like that seems like it could be valuable.
 
Having a quarantined AI system might solve many of the confidentiality/secrecy problems and totally eliminate conflict. You could have a totally separate AI which would be easy to duplicate run air gapped for total security. I'm sure there are several other creative solutions, especially with the cost of the required discrete equipment coming down. Equipment could easily be wiped or sanitized so to speak between each use, further keeping cost down. When employed over multiple pieces of equipment, the security would be near absolute. I feel a little creativity could go a long way. This is highly simplified and results could be judged on run time and the final result alone to a great extent, so the original created data would not have to looked at, keeping security near absolute.
 
count said:
I think it would be interesting if the foundries, ie TSMC, got into the EDA game and charged a wafer royalty on it as you said. Better yet, a cloud based EDA tool that could also be used for ordering after designs are validated. If it could be integrated in a sort of design to manufacturing workflow, that would be amazing. Especially for smaller customers who are designing IoT chips and are focused on time to market, something like that seems like it could be valuable.
Click to expand...

Why would a TSMC or Samsung even consider this option given Cadence’s or Synopsys’ current market caps and revenues ? Given the stock premium that an acquisition would cost, either foundry could build two leading edge fabs for the same price. I don’t know Samsung’s internal economics, but TSMC’s typical return on invested capital (ROIC) runs around 30-40%. Even though an EDA acquisition wouldn’t be “capital” per se, I’m sure that the foundries would use their ROIC as a hurdle rate for other major uses of money. Neither EDA company offers close to that rate, even before considering the revenue haircut an EDA/IP company would suffer, once tied to a single foundry.
 
KevinK said:
Why would a TSMC or Samsung even consider this option given Cadence’s or Synopsys’ current market caps and revenues ? Given the stock premium that an acquisition would cost, either foundry could build two leading edge fabs for the same price. I don’t know Samsung’s internal economics, but TSMC’s typical return on invested capital (ROIC) runs around 30-40%. Even though an EDA acquisition wouldn’t be “capital” per se, I’m sure that the foundries would use their ROIC as a hurdle rate for other major uses of money. Neither EDA company offers close to that rate, even before considering the revenue haircut an EDA/IP company would suffer, once tied to a single foundry.
Click to expand...

One word: Disruption

Do you actually think Intel Foundry or Samsung Foundry or any other IDM foundry for that matter has a chance at catching up with TSMC while playing by TSMC's rules? Much less beating them? It's not gonna happen.

Intel or Samsung could buy Cadence or make a significant investment and cut a wafer royalty deal in the cloud exclusive to their customers. Foundries, better than EDA companies, could pull of EDA in the cloud, absolutely.
 
Daniel Nenni said:
One word: Disruption

Do you actually think Intel Foundry or Samsung Foundry or any other IDM foundry for that matter has a chance at catching up with TSMC while playing by TSMC's rules? Much less beating them? It's not gonna happen.

Intel or Samsung could buy Cadence or make a significant investment and cut a wafer royalty deal in the cloud exclusive to their customers. Foundries, better than EDA companies, could pull of EDA in the cloud, absolutely.
Click to expand...

Agreed. If Samsung or Intel (or maybe even Globalfoundries if the price of oil was nearing 120$/barrel) decided it was absolutely necessary to be the foundry leader to maintain their existing mainline semiconductor businesses, I could see them pushing the "disruption" button like this.
 
All the more reason for those who are not gorillas to figure out how to make their own 80% open and effective. Arguably no one except Apple has the scale to be closed, and even they may have to change since they can't invent everything significant. Like IBM in the 70s or Microsoft in the 90s, dominance is a phase, not an institution.
 
Quarantine means fractions of total data which means inferior machine learning. There are some cases where progress has been made in creating inference structures which learn fast from smaller data sets but these, so far, have been cheating by studying the biological networks and end up working primarily for vision. The EDA area would most likely be one where the most effective learning comes from pooling the most data.
 
I was working somewhere where running everything remotely in the company cloud was an option. I gave it a shot. It worked pretty good, until the day when the AEs on the other end of the office fired up the remote conferencing software and my schematic editor effectively froze! It should go without saying, if your tools are in the cloud, when the network goes down, you are cut off from your cloud and your tools.

Now, you mention running processes in a VM. Well, sure, but do too much of that and, next thing you know, it's really a local application.

As for keeping things perfectly up-to-date all the time, I have yet to work on a project where we would upgrade the tools in the middle of the project unless we absolutely had to for a must-have bug fix or feature update. EDA tools don't tend to get the kind of QA that would allow reliably smooth constant updates in my experience and I'm sure a lot of designers would agree.

And, as a personal anecdote, we're looking at changing tools at work right now and I was discussing with my colleague how much of a shame it is that we had to do it at all, since schematic editors have only ever had to do so much without a lot of change over time. About the only additional feature in recent years that I would personally describe as a winner is version control, and I think I'm in the minority with that opinion. This colleague also pointed out the major disadvantage we have as a design company, dealing with EDA companies, locking our IP into their proprietary formats with little in the way of porting options. I can back up that assertion, having worked at a company that changed tools many times, yet still had to dig up old designs a lot. We always had this ugly step of turning a PDF of a schematic into a schematic in the current tool's database. Confining EDA to someone else's cloud would just amplify this issue. If the big boys tried to strong arm us into a cloud, I'd probably suggest to the boss putting up a few bounties on features in GEDA or QUCS or somesuch.
 
You must log in or register to reply here.
Back
Top