What Would Joe Do?
Peggy Aycinena is a freelance journalist and Editor of EDA Confidential at www.aycinena.com. She can be reached at peggy at aycinena dot com.
Crowd sourcing Design: The Panel that won’t be at DAC 2017
June 14th, 2017 by Peggy Aycinena
Why was that? Is the idea of crowd sourcing chip design a tad too open source-ish for the EDA establishment, too community based and innovative? Who knows.
The panel discussion took place, nonetheless, several weeks ago and is available below. It’s a conversation between eFabless Co-founder & CTO Mohamed Kassem and TopCoder Co-founder Jack Hughes, now Director of Tongal and member of the eFabless Board.
Per the eFabless website, the company “applies collective and multidisciplinary community knowledge to all aspects of semiconductor product development.”
Per the TopCoder website, this company has a “community of over 1,000,000 design and technology experts [providing] on-demand capability, bandwidth, and velocity so you can do more.”
The dialog below reflects both Jack Hughes’ and Mohamed Kassem’s deep knowledge around the issues of building design communities, open-source technology, and crowd sourcing design.
Jack Hughes: I sold TopCoder more than 3 years ago, but I can say that at the time a relatively small percentage of the community were engaged in chip design. Across the entire platform it was probably in the 20-percent range, but much of that group [were only engaged] in learning.
WWJD: What about the proprietary nature of chip design, how are those concerns addressed when the project is spread across so many people?
Mohamed Kassem: On our platform there are conventions, so that IP is used in a black-box approach where you don’t see what’s inside unless there’s a model.
This allows the IP – whether for a chip or a system – to be explored without the potential user or consumer seeing inside. From a developer’s perspective, the design is not visible unless the developer chooses to do so.
Is it an iron-class shield? No one can actually say there’s ever a zero-percent risk in that regard, but there is a process. It’s about knowing how to be secure [through] encryption, isolation, and data protection – all procedures we’ve learned from the IT industry. We are using strong versions of these protocols.
Yes, there’s always a risk – even Cloud providers will admit that – but we have a process in place and we know it follows a correct structure.
Jack Hughes: One of the things we found in software – particularly at TopCoder where we had the same issue – is that the customer is always concerned about security.
The business thought process goes like this: For things that are really unique, really novel, these can be protected through a patent process. But for things that are innovation-based, we saw – and still do see – a real trend towards openness.
It’s always a trade-off: How open do we want to be, versus how closed?
Look at Google and Facebook. They’re literally at the point where they will open up the entire tool set to the developers, the public. The theory is that these companies could get some competitive advantage from staying closed, but the benefits are tremendous of far more people are using the platform if it’s open. But this is all software.
As yet, hardware is still mostly lacking that [point of view]. Although, we are seeing some initiatives towards open-source hardware narrowing that domain somewhat.
WWJD: When it comes to open source, people frequently question the business model: Where’s the revenue stream for TopCoder or eFabless if everything is open?
Jack Hughes: Just to be clear, there is no link right now between TopCoder and eFabless. Mohamed looked us up and thought [our processes] in the software space could be applied to the hardware space. It’s something he has been trying to pursue.
Mohamed Kassem: Basically, I saw some similarities between the methods and approaches of the two companies, and thought these things could be applied in the hardware space.
Of course, nothing applies 100-percent, but for right now we are considering the possibilities.
WWJD: If I’m a chip design house, how can my human resources ever compete with your potential pool of designers if you’re drawing on thousands of engineers through crowd-sourcing?
Mohamed Kassem: We are not trying to get into competition with the existing [chip design] market. We are trying to compete with outsourcing.
We are working on a medium through which the community can collaborate together, to work on something that cannot be solved with current business models or current technology. We want to take the complex problems and, applying multi-domain knowledge, have the community cross-pollinate to find the solutions.
Of course, that wouldn’t be feasible without risk sharing – and it takes time to solve a problem – but to someone [willing to partner with us], we would share the resultant revenue.
WWJD: Does crowd-sourced design provide a new comfort zone compared to the limitations of in-house or out-sourced access to talent?
Jack Hughes: Fundamentally, these platforms are innovation platforms.
For instance, the Top Coder platform – from an innovation perspective, what is unique here is having access to a diversity of expertise.
There are actually academic papers out there [extolling] the extreme value outcomes of this strategy. In a traditional [commercial] organization – no matter how large – you never have access to enough expertise. This changes all of that.
In TopCoder’s case, they looked at algorithms – optimization algorithms – and found they could get results that were hundreds to thousands of times faster. Achieving these results are fundamentally what these platforms are all about.
Companies looking for their next increase in performance, something they are usually grappling with inside of a bigger problem.
WWJD: Given the numbers you suggest are available through crowd-sourcing, is the established model of funding a professor and their grad students to research specific problems now dead?
Jack Hughes: It’s not dead, but what we are talking about changes it. Rather than having the really good luck of getting a generally small group of people in a lab to come up with some insights, the group is now much, much larger.
There was a Harvard study that used crowd sourcing to solve a software problem. The person who came up with the solution – and there were hundreds submitted – worked in Italy and was not even a computer scientist. And that suggests that someone who has a totally different perspective on a problem can have a particularly successful insight.
I wouldn’t say that the lab at the university is [no longer relevant] – they are still very good at solving thorny problems – but even they will begin to use these tools as a way to reach that unknown person with a different perspective.
Mohamed Kassem: And one important component you have to remember: In order to fund a company to profitability, you must have an internal team who can build these relationships at the university. This is costly and time-consuming.
It’s a completely different approach to use an open community, using the Internet, where you’re not selecting the team. Any university anywhere can apply their discipline to solving a problem, but our strategy expands on that.
We are adding the [strength] of crowd sourcing to chip development by building a community. Using our knowledge and engineering expertise, we are going to be able to solver bigger and bigger problems.
WWJD: What do you say to a customer about the risks involved in sending a design out to the crowd?
Mohamed Kassem: Actually, we say there are advantages and benefits from working with a community. And these may even reduce the risks – the time and costs – especially if the problem is something new, something at the IP-level.
For instance, say I want a new type of design that has never been done before – having a community approach with multiple paths to the solution actually de-risks that project, particularly when time to market is part of the consideration.
WWJD: Still, how can you reassure me, as a customer, that my design is safe from IP theft during the development?
Mohamed Kassem: The customer is 100-percent aware of what is going on. And as always, when it’s a community engineering project we apply our security protocols.
Some parts of the community may not even know what the larger project is they are involved with – it all depends on the problem being solved.
Jack Hughes: Mohamed, maybe you could walk through how an eFabless project works.
Mohamed Kassem: A specific project would be one we did for X-FAB.
The company had a set of specifications and a request for IP with certain requirements. We captured those requirements on a data sheet – the specs were not secret – and published it online on our website.
After it was posted, 88 different designers accepted the challenge, and we carefully went through the process of reading to each of them the terms of accepting the assignment.
During the course of the challenge – there were actually 23 designers who were actively working on the project – we didn’t see their work, but saw the outcome. By the end, 6 designers fully completed the design to specification.
Two of them had a different architecture, and two were backed by companies – meaning a company was supporting those designers. It was a very open process. The designers did not see each other, and the outcome was measured against the specifications.
It’s true, there are cases where the customer says: The specification itself is secret, perhaps even locked in a patent. Then a trade-off has to be made between being closed and the advantages of openness. The customer has to see the advantages to [move forward].
Also, there are cases where we can partition the project into segments, where it [would be impossible] to converge in on the whole system under design. But there are also disadvantages here: Who owns the partitioning? The community?
Jack Hughes: These are fairly common issues. Security in IP is not unique to these [situations], even within a company these issues can come up.
Recall that Google right now is going through a big thing with some engineers who left to do a startup. They took their knowledge with them, and [now there is a possible problem]. These issues are not unique to crowd sourcing.
For the types of platforms [we are talking about] to succeed, they have to be process-oriented, which also makes it a bit easier to manage security. You can capture the specs and the data sheets and be very specific, then the platform has to have a very accurate document that [details] who contributed what to what.
In the case of X-FAB that Mohamed mentioned, they were looking for an improvement to an existing design and the platform made that doable.
Mohamed Kassem: Yes, X-FAB wanted to try the model all together, to compare apples and apples. They were surprised in a positive way by the size of the engagement from the community, and the quality of the outcome.
WWJD: So why isn’t this an important topic be discussed at DAC?
Jack Hughes: It’s not a conspiracy: We ran into these problems at TopCoder all the time.
If there’s anything that’s new or innovative, or suggests change, there is always a tremendous amount of resistance. People get used to established approaches and structures, but if the change brings benefits, ultimately it will win out.
When something comes along that is truly disruptive, people have no choice but to engage with the innovation. Yes, it would be better if they were a little more open-minded, but that’s not human nature.
Mohamed Kassem: I also wouldn’t say there is a conspiracy.
But as Jack said, there’s a tremendous amount of resistance to anything that requires change. There are risks. And added to that mix, specifically in the semiconductor industry, we are bound by a lot of constraints.
The expression ‘open-source’ does not sit well with the foundries, for instance. It takes a lot of explaining, and from the many different angles, to convince them that both the changes and the open source can be managed.
It does sometimes seem that if you work with the community, the foundries will never work with you. But we feel that door may be opening. We participated in a panel recently, and afterwards a company [in the audience] came to us to contribute to our platform and tools.
When the EDA providers see this community, and the potential benefit to them from working this way, they will see [the advantages] versus their traditional closed platforms.
WWJD: So going back to an earlier question, this crowd sourcing and open-source mentality leads to the business model being more about services and not about products?
Jack Hughes: It’s not such a big change. All we are doing is recognizing the reality of the design environment going forward.
My whole background is in software, and I know the last 30 years have been about nothing but a march toward open source. We wouldn’t be able to use the products we need today without open source, including the phones we are speaking on.
The software world has accepted it, even Apple – their OS is based on Linux. You wouldn’t have companies like Facebook or Google without open source, and many of these companies have become very large.
This is going to happen in hardware as well, although for now it does make people nervous. Nonetheless, I think people are more nervous than conspiratorial.
They do see that the semiconductor industry is in a tough place right now, having a more and more difficult time growing [because of the costs]. That is always an indication that it’s time for the next wave of adoption.
So then the question is: Who has the scale to build on the kind of open source design we are talking about here.
Mohamed Kassem: I call it mixed-domain knowledge. Open source uses existing resources to jump-start innovation and standardize things.
Today, as Jack said, the entire software industry is built on open source. It’s very easy to take the Linux platform and use it in a reliable way, and it was developed as open source.
And this will be the case eventually for hardware. [although at a slower pace] with the big companies resisting change.
But look at the real innovation happening today at companies like GoPro or Fitbit. These companies are small, and the world is going to see a lot more of them.
Jack Hughes: Actually there are also hardware examples.
Take a company like Apple – for all intents and purposes is was a computer company and not doing well. Then they instituted change across the organization to absorb the next big thing, which was PCI. Without that change, Apple could not have created the iPhone.
WWJD: In the last 90’s, I asked Wally Rhines if Mentor would ever support Linux and he just laughed. Certainly that attitude changed within a brief number of years.
Mohamed Kassem: I come from the semiconductor industry, and the EDA industry always follows the semiconductor industry.
There were EDA companies, for instance, who tried to use Cloud computing for verification, but the answer was always: Our customers don’t feel comfortable with it. The EDA companies didn’t have problem with the concept, but their customers did.
The semiconductor companies want things done a certain way, and the big three EDA companies will not hurt their existing business models by challenging those customers.
But at the end of the day, all of these companies must be open to change. Yes, it’s a long-standing industry, so change will be hard.
Jack Hughes: Change is hard, but the alternative is worse. The companies that drive change, those that are in so much earlier than everyone else, they are the ones who succeed.
WWJD: Which brings us to the received wisdom that small companies drive innovation, while large companies acquire small companies as their only access to innovation
Jack Hughes: Yes, for large companies acquisition is Plan B for change. If you can’t drive change through your own company’s DNA, then you’ve got to acquire it.
But even then, it’s tough to do. Every acquisition is a challenge
WWJD: Speaking of open source, the RISC-V movement is showing an ever-quickening pace of acceptance. Isn’t this an endorsement for crowd sourcing?
Jack Hughes: Yes, RISC-V will be a boon to the industry, if you can get some of the larger companies to adopt it. Then the volume of [implementation] alone will make it successful.
But asking smaller companies to launch these types of products is too difficult and too expensive in the current environment, even though directionally open source is where we think things are going.
Mohamed Kassem: Just look at RISC-V and see who writes about it: Google, Nividia, and so on.
The RISC-V organization has open-sourced the instruction set, and a lot of people are now building various architectures that are under test. Not just a single architecture, but multiple architectures. RISC-V is an open platform that support this.
But this is not an open-source ARM. All of these architectures still lack the analog interfaces, so this is another opportunity: Get around that limitation and build something complementary to it.
Jack Hughes: We philosophically believe in an open ecosystem. Microsoft did not benefit from keeping theirs closed.
Since the RISC-V instruction set is published, nobody gets to go in and say: Choose us because our version is best.
Mohamed Kassem: It’s not exactly open source, but look at GitHub. There are thousands of projects being built on [their platform], with hundreds being very mature and complete. And there is hardware based on it, with completely open source code.
And Amazon has launched a platform for FPGA design, uploaded to the Cloud, that you can access for just $10 an hour.
WWJD: It does seem with everything happening so fast, if you blink you’ll miss something and fall behind.
Jack Hughes: Yes, open source happened very fast in software, but there’s a lot that goes into making a hardware device.
And as they become more automated – more pluggable – it’s the technology driving itself to a certain extent. Things are opening up, and we believe this is a good thing.
Mohamed Kassem: And now there is a company which is a foundry for 3D designs. You bring your design to them, choose your materials, and they manufacture it and send it back to you.
The established manufacturing houses will continue to have a role [in the semiconductor industry], but as the designs become wildly open, things will change there as well.
Jack Hughes: The Amazon Cloud is now supporting thousands and thousands of companies, all figuring out the way to add value to the success of open source design.
The people who will be successful will add value into that open source equation. They will figure out how to add value on top of what is open and available.
WWJD: This all seems very exhilarating, although it might take a long time?
Jack Hughes: How long will it all take? Again change is hard, and there is a natural aversion to change.
Some of these companies have good reason to stick to what they’ve done and we’re sympathetic. What we are proposing is not a conspiracy, it is something positive.
Mohamed Kassem: I think about the EDA companies as potential collaborators and partners. while our focus is on enabling our community. We are open to hearing about how to make more of the EDA technology available to that community.
Jack Hughes: We want to help the EDA companies – that’s our pitch when we talk to them.
We are not trying to compete directly with them, many of them have years and years of functionality and institutional knowledge. But ultimately, we want to work with them and try to help move the technology much more into an open environment.
WWJD: Is it hard to describe the future?
Jack Hughes: We don’t necessarily not know the future: There will definitely be a business model pivot – that’s always a part of change. People and industries have to change, and we will need to feel our way through the process.
Mohamed Kassem: We certainly believe that the future will be this way, with crowd sourcing and open source. And because this is my conviction, my belief and the team at eFabless, we are putting a lot of energy towards getting us there.
That doesn’t mean it will actually happen, but it’s always belief that gives you the drive to do the difficult thing. And we are certainly learning through this journey.
Tags: Amazon, Apple, ARM, Crowd sourcing, DAC 2017, eFabless, Facebook, Fitbit, GitHub, Google, GoPro, Jack Hughes, Linux, mixed-domain knowledge, Mohamed Kassem, NVIDIA, RISC-V, Tongal, TopCoder, Wally Rhines, X-Fab