New Market Drivers

Experts at the Table, part 1: The industry is changing. Who is driving the market today and what new requirements do they have?


Semiconductor Engineering sat down to discuss changing market dynamics with Steve Mensor, vice president of marketing for Achronix; Apurva Kalia, vice president of R&D in the System and Verification group of Cadence; Mohamed Kassem, CTO for efabless; Matthew Ballance, product engineer and technologist at Mentor, a Siemens Business; Tom Anderson, technical marketing consultant for OneSpin Solutions; and Andrew Dauman, vice president of engineering for Tortuga Logic. What follows are excerpts of that conversation.

SE: Semiconductor design and manufacturing have been driven by the mobile phone industry for the past 10 years, but that is changing. How much influence do mobile phones still have and what new drivers are coming into the market?

Kalia: It is true that mobile phones have driven a lot of the design for the past 20 years, and that kind of momentum will take some time to slow down. So mobile is still driving a lot of the development, but something interesting is happening. If you look at an application such as ADAS, the sheer complexity of that application makes it conducive to taking the learning from the mobile platforms and migrating them. While it is not a continuum, there is a lot of technology, design methodology and verification methodology that has become formalized and matured and these can be applied in the automotive space.

Kassem: We spent a lot of time consolidating features into a phone and making it as small and low-power as possible, and now we need to dismantle it into different pieces with different radios and smaller purpose. The phone is the ideal IoT device. If you could shrink it to a dot, then you are done. But we can’t, so it has to be made smaller.

Dauman: Mobile drove everything. It drove the implementation, it drove verification, the process nodes and the need for low power. As we break it up for each market segment, not every piece will be driven as much or be as important as it was for mobile. Nothing will have quite the impact that mobile had, not the volumes.

Ballance: And what you care about is changing. Self-driving cars require stability, full verification and concerns about the safety critical nature of the design.

Kalia: Mobile has a different volume level than automotive, but it may be worthwhile talking about the nebulous concept of IoT. From a volume point of view, that may be even higher than mobile.

Kassem: You have to consider the market. If you look at books, there is a long tail. There are best sellers and then you have a very long tail where they may only sell a thousand copies. The aggregation of that low volume is such that you can’t afford to develop products at that low scale. If you cannot afford to develop products with low enough volume, that are still profitable, then there is a lot of risk. You do not know what the next killer app is. Look at the number of products on Amazon. Which one is better?

Anderson: Products have to differentiate.

Kassem: It could be that one has a better battery life and doesn’t need charging as much. The market will determine the volume and the best sellers will emerge. GoPro was founded by a surfer and went from the long tail up to the best sellers.

Ballance: What drives the long tail is technology that can be re-used and re-purposed. With mobile you are developing a product, with IoT you are developing technology that will get remixed in many different ways and you don’t know how.

Mensor: FPGAs are very much in the long tail. For the past 15 years there has been a fairly large segment of FPGAs that are driving volume, and the most notable is wireless. These are driving into the millions with reasonable ASPs. And in the data center, we are seeing Project Catapult become successful. But if you use that as a proxy, then you see a programmable hardware tool for the long tail. What it takes for programmable logic to get to higher volume is a confluence of power and cost. And while flexibility is beneficial, it has to be weighed against the realities of something that would make sense from a consumer point of view. With FPGAs you can go to four digits and even five digits for the price of the product. On the other end of the scale are embedded FPGAs, and now we are talking die sizes that are a fraction of the entire FPGA because you don’t have to build an entire infrastructure. You just build in what you want. How does that play in IoT? It will play in cellular handsets, at least in a prototype domain, but less certain for a mainstream domain. And in IoT it depends. At some point you look at programmable, sequential solutions – CPUs – and say that is fine. Your programmable logic offers a value proposition that is specific.

Anderson: There is a tendency to dismiss IoT as being something really simple, which involves parts that are thrown away. But that is not reality. I am seeing people with sophisticated designs, including many with high-end FPGAs. You may say you do not use an x86, instead going to an Atom processor. But when you look at that datasheet you see eight cores and a server-class chip. This is a serious design.

Kalia: One of the reasons why IoT has been stuck with that perception is that while an application like mobile, and even ADAS, have gone to the level of designing a complete system, IoT has remained at the level of devices. And then it is left to aggregators, the solution providers, to take these off-the-shelf devices and create a solution.

Kassem: Everything existed off-the-shelf before the emergence of smart phones. Every custom component was driven by smart phones, including screens and batteries. When you get into specific applications, you really optimize. You go from standards parts to an ASIC perhaps. For some devices you may not need synchronous logic, instead going with asynchronous logic that will be turned on and off in bursts. This is customized for very low power.

Anderson: There are IoT devices that are nothing more than simple sensors, but that is the brush that is swept over the whole industry.

Kassem: We have to make them profitable, and the answer for that is a platform technology.

Dauman: There is a perception that IoT devices are small, but you just have to look at self-driving cars and this is a very large, heavy, high-risk, high-impact IoT device. It has sensors, it will be communicating with its neighbors.

Kalia: It will be the most complex devices that exist. They are not small or simple.

Anderson: Right. Automotive is a part of IoT.

Kassem: Except I don’t think there will be 3 billion cars.

Anderson: But it is bringing the biggest evolution, which is safety. People who build pacemakers and missiles had to think about it a long time ago and build high reliability into them, worry about alpha particles, etc. All of a sudden, that is a part of daily life. With a self-driving car you have to be sure that is just as robust as pacemaker.

Kalia: What has happened because of safety in cars is that the economics has evolved. When you talk about a missile or pacemaker, traditionally there was a simple way of ensuring safety – redundancy. You just put in two of everything ,and if you have a missile costing $10M or $100M, putting in two copies of the hardware doesn’t impact the cost that much. With a pacemaker, that would change price from $200 to $400 by adding redundancy. This is also not too big of a problem. But if you take a $20,000 car, where the electronics costs $5,000, and you want to add redundancy, that may push it to $8,000. That is a noticeable difference. This creates new design paradigms. How do you ensure safety without resorting to redundancy?

Kassem: The electronics in the car didn’t fundamentally change until we started to add the sensory systems.

Dauman: Safety used to be about fault-tolerant, triple-modular redundancy. In fact, we used to talk about security and safety as being different domains, but they can’t be because you can build in all of the redundancy you want and if someone gets in there and changes the function, then fault tolerance and safety are gone. So it has become one discipline. We are not doing this well today.

Kassem: There are multiple disciplines involved. When designing a chipset you are relatively contained. But when designing a vertical it may require medical knowledge, 3D design, electronic design – all of these together. Being able to bring multiple disciplines together is important.

Anderson: You have a vertical, multi-discipline thing going on, but you also have horizontals including verification and safety and security – that makes a complex matrix.

Kalia: Bring it back to autonomous driving, like we saw with mobile 15 years ago, the complexity of the problem is far ahead of the ability of the people to grapple with the complexity. The people trying to solve these problems are still at an early stage of wrapping their head around the field.

Kassem: There is a lot of trial and error. You have to keep experimenting and learning.

Kalia: There is a transition from mobile to these kinds of applications. All of the designers who have had success in the mobile space are assuming, mistakenly in my opinion, that it will be an easy transition. I have had many conversations with people who have a chipset that was excellent for mobile, and they want to transform it into an ADAS system.


Gordon Walsh says:

“I have had many conversations with people who have a chipset that was excellent for mobile, and they want to transform it into an ADAS system.”

Based on what I’ve experienced I would say its more like “I have a chipset that was excellent for X and I transformed it into an Embedded system and I now want to transform it into an ADAS system.” – while possible meeting the safety requirements with such an approach usually implies pushing the safety functionality onto support components outside the chipset – the amount of complexity that causes is not for the faint of heart!

Anand Shirahatti says:

Summary for those who are short on time: Keywords: Mobile, IoT
[1] Mobile drove the technology nodes, implementation, power and verification
[2] Now there are other interesting areas emerging as well. Need to deploy mobile leanings there
[3] Mobile => Product; IoT => Technology
[4] IoT has gotten over simplified due to few devices. In reality, IoT Spectrum ranges from sensors to server class chips with multiple cores
[5] Classic complex IoT example is self driving cars with complex sensors and talking to neighbors
[6] Self driving cars – How to balance cost and safety without excessively relying on redundancy?
[7] Safety and Security should not be viewed in total isolation. More holistic approach is required combining multiple disciplines to build successful solutions

Leave a Reply

(Note: This name will be displayed publicly)