EDA и ESL в автомобильной электронике
Jeff Roane, вице-президент по маркетингу компании VaST Systems Inc., Marios Zenios, бывший старший вице-президент Motorola Automotive Group, в настоящее время главный управляющий Altos Consulting, Dr. George Saikalis, директор по разработке (R&D) автомобильной электроники из Hitachi America Ltd обсуждают с корреспондентом Electronic News/Electronic Business возможности system-level design (проектирования системного уровня) по результатам Конференции по автомобильной электронике (2007 Design Automation Conference), прошедшей в начале июня в Сан Диего, Калифорния. Беседа публикуется в сокращенном виде
: What is required in the supply chain to reduce the number of software problems in vehicles?
Saikalis: There is a tremendous amount of software being used between control algorithms and diagnostics in vehicles. You need advanced tools to be able to not only allow you to come up with your control algorithm but to make sure all the test cases are covered in terms of I/O. Verification is very important.
Roane: Flashback to the panel that we just left [at the Design Automation Conference], and Frank Winters at Delphi made the comparison and actually kind of commended this industry for advancing verification for IC design. And he made the point that the analogies are pretty clear. Today, it’s typical for companies to bring to market multimillion gate designs. In excess of 10 million gates and have those design work, and work flawlessly. And it’s because of the advancements in verification technology. I began my career at TI back in 1985 and we were doing a design of 10,000 gates. I spent my early career there in a lab with a big TTL emulator of an IC. We basically replicated the IC in hardware using TTL devices for the purpose of verifying the device. It took forever and then all of a sudden simulation technology got introduced and made it possible to simulate the entire IC and do a much better job of visibility into the operation of the IC and then controllability so you can induce errors to see how the system responded to those errors. The exact same thing needs to happen for software. I think part of the solution is simulation/virtualization – you could almost use those terms interchangeably with the same concept.
Saikalis: We’re talking about, say, one controller in a vehicle, but there are many controllers now: body control, chassis control, engine control, telematics control; it’s unbelievable the amount of software size.
Q: With 100 chips in, for example, a General Motors’ high-end vehicle, the system is so intricate in terms of all the systems interoperating, doesn’t that requires a tremendous software effort?
Saikalis: The thing about hybrid car systems is that have both electrical power train, plus the internal combustion engine controls running together: how do you control both properly?
Zenios: There is no question about the growth and the evolution of electronics that consumers are going to demand out of their automobile. So the challenge is to bring this industry [the electronic design automation industry] closer to the automotive industry.
Roane: The other factor we are seeing play out is that these features that are delivered through software and electronic systems are in high demand from customers. So either the automotive OEMs will figure out how to deliver those solutions, or third parties will deliver them. I think the OEMs have woken up to the fact that they are losing a lot of money to third party, add on navigation systems, entertainment systems, cell phones -- so they are focused on delivering the same, so the vehicle rolls off the assembly line and its got everything included. As to whether any particular OEM wins or loses, it will be based on their ability to adapt and develop and deliver high quality systems that meet or exceed the quality expectations of the end customer.
Zenios: Jeff is bringing up this other challenge that now you’ve got electronics on the vehicle interacting with networks and the need for tools to do that grows exponentially because of the use cases and the unknowns that that environment will create. For example, you’ve got simple things like iPod integration into the vehicle where you are bringing in an external device that you don’t know the level of quality of the software – which we do, by the way, with the iPod – and you are plugging it into the electrical system of the automobile. So now you’ve got a more complicated interaction. With telematics systems you are accessing the Internet now – you are bringing data into the vehicle.
Q: For advanced, next-generation automotive features such as collision avoidance, what is required of software to make it ubiquitous?
Zenios: A lot of my experience has been being essentially a pioneer in the telematics arena. The way I made sure there was quality in the software is that we spent a lot of money and resources testing it. I wish I had some automated systems. Just for the verification of that software, such as for one telematics platform, like [General Motors’] OnStar, you’re talking in the neighborhood of $30 million for the development cost and 600 software engineers around the world working full time on this. But we cannot afford to continue to spend at that rate. We have to figure out a way to automate the verification and that’s going to allow us to add more functionality.
Q: How do you ensure you are delivering applications for mission-critical tasks that work flawlessly all the time?
Roane: The way these applications are made to work today, there are tremendous dollars thrown into testing to make sure the software works flawlessly. That is not sustainable because as complexity grows, you can’t afford to just throw more bodies and resources to make that software flawless. So, this brings opportunities for automation and opportunities for the [EDA] industry.
Q: How do you determine what specific tools are needed?
Zenios: The way it should work is that the EDA players have to get closer to understand the needs in a productive sort of way. If you wait for the customer, sometimes the customer is too busy fighting their own wars, to sit down and define. What this industry needs is to dedicate resources to get closer to the car manufacturers and the Tier Ones to sit down and understand not only the component level but the subsystem and the entire system. And then step away from that and look at what you are doing and try to fine tune it. Try to develop additional technology that would meet those needs. That would be my recommendation on how to get this done.
Q: Are the car manufacturers and Tier One suppliers willing to pay for those tools, and to what extent?
Zenios: They are today. The better it is, the more money they pay. [much laugher ensued]
Q: Are they willing to help develop these tools?
Roane: The short answer is yes, although there are some fundamental differences in technology adoption or product adoption patterns and cycles and times for the automotive industry vs. the semiconductor industry, for example. This is not an industry that has been covered or addressed by EDA so all those lessons learned don’t apply: there is a fundamental shift in being able to comprehend what’s important to the end customer. “Time to market” was the mantra for the semiconductor industry. All that mattered was how quickly you could get that chip done and out to the customers. For automotive, everything we’ve been talking about so far really comes under the quality umbrella. If you can convince the customer that your solutions do indeed improve quality, which has to be measurable, then they will pay.
Saikalis: It boils down to cutting down development cost. Instead of having many iterations of the hardware design, you do it in a couple, depending on the specific application. That’s very important. Also, sometimes if you talk to a certain OEM, they may not have the vehicle available to test your system, so what do you do? You have to wait closer to pre-production or just do it in this kind of environment.
Roane: The automotive supply chain is complicated and it is insufficient to show up with just a processor when the customer is trying to simulate a power train or suspension subsystem. You have to have all the components of the solution in order to effectively make the sale or make the case that your technology is worth displacing whatever was there before you got there. To do that, you have to work with the other members of the supply chain.
Q: Is it realistic to think that an EDA company could play in this industry?
Zenios: You need a good go-to-market strategy for the industry towards the automotive industry. That doesn’t exist today. Point solutions don’t necessarily work. This is a highly custom-intimate environment and it takes a big, different go-to-market approach. It involves a highly complex sales strategy because if you are selling point solution, the selling of that has certain challenges in terms of how do you sell a point solution which includes the reputation of the company, the attributes of the product, the quality and price of the product, and you go sell it. When you get into this highly complex space though, all those things are true but in addition, now you are dealing with a whole bunch of different people at the customer than you dealt with before with your point solution. You have to build additional relationships. Then you have to architect – it’s upon you as the EDA provider – a vision of what you are trying to do, and also be able to articulate a high level architecture of the whole system, which is the end point. Now you are creating an environment that you want to play in higher up the food chain. Once you do that, you have to facilitate technical people to technical people. That’s how you get the sale.
Q: Where the EDA industry is today, will it be possible to do what Zenios described to play in the automotive space?
Roane: What it will take for EDA to shift focus will be a realization that there is opportunity but in playing in that, we have to detach ourselves from the silicon piece. We [VaST] certainly work with semiconductor suppliers primarily to provide those models when we show up at a Hitachi, or Delphi or GM, that we have all the models of the components of their system that allow them to do what they want to do.