EVs Raise Energy, Power, And Thermal IC Design Challenges

The transition to electric vehicles is putting pressure on power grids to produce more energy and on vehicles to use that energy much more efficiently, creating a gargantuan set of challenges that will affect every segment of the automotive world, the infrastructure that supports it, and the chips that are required to make all of this work.

From a semiconductor standpoint, improvements in thermal management will be needed to prevent chips and subsystems from overheating as they process more data from more sensors, and communicate with each other and the outside world. Hardware and software will need to be designed together, so different functions can be partitioned and prioritized to improve efficiency. And there will need to be changes in battery management and battery chemistries, as well as improvements in aerodynamics.

“There is an entire paradigm shift happening in that everybody’s thinking power,” observed Preeti Gupta, director of product management at Ansys. “Gone are the days where it was about performance and area. Power is such a critical design metric now. You’re hearing about performance per watt. It’s not just about performance alone, but at what cost of power are you delivering that performance. The automotive industry is very interested in understanding power and thermal impact early in the design flow. They are making huge decisions in terms of their design choices. When it comes to chips, it’s a lot about the packaging, not just what is on the chip and burning power. Like mobile or handheld device developers — which were concerned about power and thermal footprints because it should not burn your hand when you’re holding a mobile device, along with battery life considerations — when it comes to electric vehicles, the same battery life extension applies. Thermal aspects are even more important to model early in order to make the right design decisions. At a high level, everybody is moving to shift left, which is getting feedback as early in the design flow as possible.”

Going forward, EV architectures will require many more compute operations as vehicles are increasingly digitalized. As a result, different architectures are evolving to meet EV vehicle compute demands.

“One approach is a centralized compute domain, which is probably the ultimate architecture being discussed, and the automotive ecosystem is trending toward that architecture,” said Ramesh Chettuvetty, general manager of memory solutions and RAM at Infineon Technologies. “We’re not there yet. We are still in the domain/zonal architecture phase, which is a distributed domain approach. However, once we have so much compute — and autonomous driving picks up and there’s all kinds of sensor data coming in — the processing workload is definitely going to be massively high. This will increase the thermal dissipation of these compute elements.”

These factors combined also increase design complexity, because chips need to be designed as parts of systems, or in the context of systems of systems.

“It is already challenging to design semiconductors to work on a broad temperature range environment,” Chettuvetty said. “Most of it peaks at 125° Celsius. If the thermal dissipation is much higher, as part of the need for higher compute, there will definitely be challenges like what we are facing in data centers and other applications that require coolers or fans. Those are additional overheads that will consume some energy because it’s all ultimately going to run on the EV battery. This will bring down the mileage of the cars, which is one of the key features that OEMs are going to promote going forward, so they’ll definitely want innovative architectures to get around that problem. Power is the bottleneck everywhere, regardless of whether it is wall-powered or battery-operated. All engineering teams must move away from traditional ways of implementing features and look to innovative ways to do things.”

Still, he expects automotive systems architects to start adopting innovative solutions to address these problems, and explore the alternatives that are available. “Designs should be approached in a comprehensive way. The engineering team has this habit of looking at it in a very limited scope. We have to put all those pieces together, and somebody has to take a comprehensive look at it to see whether all the assumptions that are being made by the engineering team are realistic in a practical world.”

This represents a fundamental shift. Traditionally, the way the hardware engineers thought about designing these chips was to put in many operational modes.

“They’d say, ‘I can turn down certain things or I can monitor things, and I can use that monitoring to meter what I’m doing.’ Maybe I slow things down in one area or another,” said Steven Woo, fellow and distinguished inventor at Rambus. “However, what we see more of in AI, which is likely to play out in all fields, is that the software guys really understand quite well the tradeoff between the performance of the system and the precision of the system. The way they think about it is if they are somehow limited in bandwidth, energy, or something else, they turn it into a software problem. ‘If I need more bandwidth, then I can reduce the precision of my numbers. Instead of 32-bit floating point, I do 16-bit floating point.’ Then in the bandwidth, they can get twice as many numbers. They do give up precision for this, but they know how to deal with that. As a result, they train the AI algorithm specifically for reduced precision or for sparsity.”

In the AI arena, there is more of a holistic view for integrating hardware and software, Woo said. “In the same way over the last 20 years, programmers have been forced to become more architecturally aware. What size cache do I have? What exactly is the architecture of the processor I’m running on? Programmers will have to be more cognizant about things like power limitations in the system, and thus try to use tools and APIs that let them trade off power for performance. That’s how I believe the evolution will happen. It will take time, because it’s not easy to think about these things. It’s taken about a generation of programmers to really understand that you can’t be as abstracted anymore in what the architecture looks like when you’re writing your programs. It’s going to move in that direction over the next 20 years.”

Where the power goes
While this evolves, the ECUs themselves will employ advanced power management techniques and play an increasingly important role in overall power management within the architecture of an EV. A key consideration is knowing where the power is spent.

“A lot of power is spent when you transfer data,” said Sumit Vishwakarma, principal product manager in the AMS Business Unit at Siemens Digital Industries Software. “You want to minimize what you’re transferring, and only send sensible information. That’s where the ECU comes into play. The ECU is mostly digital logic, and then lots of PHYs connect that to different parts of the car. There are also a number of sensors around the car connected to the main hub integrated ECU. Other than that, there are different ECUs working separately for different applications. For example, when you are a passenger, and you sit in a passenger seat, the first thing the car seat is going to do is detect there is some weight. That means it gives the signal to the main display unit to turn on the light indicating the passenger needs to put their seatbelt on. The moment they sat, you will see that it is on, and when they take the seatbelt and buckle it in, it sends the signal to turn off the light indicating the seatbelt is fastened. All of this communication is happening every time. But when the car is parked, then those things never happen because all of the sensors are not necessary to be working at that time, compared to when you’re driving the car. This mean there is always power management happening inside the ECUs.”

The question then becomes what should be on, off, or something in between. That requires defining the power intent for the ECU, such as within UPF. “From a design perspective, the electronic control units, which are primarily digital circuits, could be implemented with some kind of system level language, such as System Verilog or other functional language or could even be implemented using FPGAs,” Vishwakarma said. “Those are primarily digital circuits, and their job is to make sure that it takes care of the sensor fusion and the decision-making.

From there, power domains can be specified to indicate different power scenarios, and how much power each block should have. In addition, power gating, isolation, and retention can be used for controlling the flow of power between blocks.

There are also a number of power considerations stemming from the movement of data in EVs. “There will be multiple networks, all chained together,” said Paul Graykowski, senior technical marketing manager at Arteris IP. “We’re putting more and more devices on a vehicle. It’s not just, ‘Drive me from Point A to Point B.’ It’s, ‘While driving me from here to here, I want my stereo. Passengers want their TV. I want my air-conditioned seats. I want my auto driving. I want the sunroof.’ There are so many little things, which means the user interface is going to be very complicated, and we must make sure we have a good flow with that. There are going to be power needs everywhere, and we’re going to have to address those power needs. It might be simple where we just say, ‘This device is going to be fine, but this other one needs a very complex network.’ We’ve got to have the timing parameters set forward as well. It’s all going to come into play.”

CFD
From an EDA tooling point of view, one tool area that formerly focused on mechanical design is now being applied in EV design. Computational fluid dynamics (CFD) are being brought to bear to improve the energy consumption profile of electric vehicles.

Robert Schweiger, director of automotive solutions at Cadence, said that based on consumer surveys, the number one concern about buying an electric vehicle besides the price is the range of the vehicle. “The range needs to be beyond 400 kilometers/248 miles, because the charging takes quite some time, and it means an additional stop on your journey. The range can be significantly improved by optimizing the aerodynamics. Therefore, CFD will play a major role in optimizing the aerodynamics of a car, which has an impact on the range.”

Heat is another consideration. With internal combustion engines, temperature can spike in traffic jams. EVs, in contrast, use less energy in stop-and-go traffic, but can overheat during rapid charging.

“The challenge for an EV is during charging,” Schweiger said, “CFD software can be also used for cooling systems to simulate the water flow in how it is able to cool down systems. For batteries, new concepts are emerging whereby, alongside the battery pack, water is sprayed from the top of the system onto the battery while using a supercharger. With a tank underneath, the liquid is pumped up, and sprayed down again. This is a new concept that will be available to cool down batteries, which in turn will eventually help to improve the lifecycle of a battery, as well as cool the battery down.”

Doing more earlier
Understanding and predicting power needs begin very early in the design cycle of automotive chips and systems. In concert with that, many EDA companies now are doing more modeling to get tighter accuracy and improved predictability, said Ansys’ Gupta. “If you’re thinking about an early design phase, and you’re talking about a much later design phase, so much of the design is implemented,” she said. “How do you predict that early at RTL? What would the placement be like? What would the wire routing be like? Modeling all of these components is now possible. There are techniques that we can employ to model these now.”

The goal for many of these design elements is lower power, and that has many elements, Gupta said. “How many power domains do you need? Can you really turn on or shut off for mission critical applications? You probably cannot have logic shutting down and waking up in some applications, so it’s all about understanding where power is going and why, where activity is and why. Once you understand the ‘why,’ then addressing it becomes easier. So it’s a lot about early power visibility, a key component of which is early power visibility from emulation use case scenarios. You can’t live within a well and just say, ‘I’ve optimized it.’ Look at the real traffic, the real application use cases. Is your device really energy-efficient and power-efficient?”

New architectural challenges
To address these numerous energy, power and thermal constraint factors for EVs requires a different way of looking at designs. “First of all, the focus on power consumption becomes more relevant,” said Tim Kogel, principal engineer for virtual prototyping at Synopsys. “With a traditional combustion engine, you basically have infinite energy. You don’t care if you burn 100 or 200 watts of power. But with electric vehicles, that all impacts the range in a much more significant way. We’ve heard that with 100 watts of more power consumption, it costs 10 to 30 kilometers of range. Vehicle range is so much a unique selling point, and range anxiety is one of the key concerns for electric vehicles.”

On the other hand, with Level 3 and Level 4 autonomous driving, there are extremely complex, advanced chips. “There are multiple cameras, multiple other types of sensors,” Kogel said. “There are complex algorithms, like classification and segmentation, pathfinding, the most complex processing tasks. When you just put this in a generic, general-purpose GPU type of architecture, it burns too much power. We’ve all seen the photos of autonomous car prototypes where the whole trunk is racks of compute that require kilowatts of power, which is in total conflict with this requirement. To get that to an acceptable level, you need to move from the general-purpose architectures to much more dedicated architectures like vector DSP, application-specific instruction set processes — or even for some pieces, hardwired logic, because only that gives you the level of power reduction by one or even two orders of magnitude that you need to have this level of functionality at this power budget.”

It’s the same for communication, which is also a big part of the architecture. “Here, the easy thing is always to use caches and cache coherent interconnect and then the data is moved around more or less automatically,” he said. But again, it’s burning large amounts of power, so people need to go to dedicated memories, dedicated memory transfers. The price that people pay for this is that these dedicated architectures are less flexible. You don’t have the easy push button software flows, so the software implementation becomes an effort to implement and to verify.”

Conclusion
Will the issues with energy, power, performance and thermal push back widespread deployment and adoption of electric vehicles? Combined with ADAS systems, which are extremely power-hungry, the challenges only seem to be growing.

Vasanth Waran, senior director of business development for automotive at Synopsys noted that while electric cars bring a new challenge because of the limited range and limited battery power, ADAS has power problems even on the internal combustion engine side. “It’s a very complex problem, and it’s as much a thermal problem as it is a power problem, because the car is one of the most challenging environments you can design a semiconductor to operate in. Everybody gets irritated if their audio doesn’t work or their map suddenly gets shaky or their camera goes off. You always want the best performance. You still want the A/C to work. You still want the cameras to work. It’s a hard problem to crack. And because of this, over the last couple of years the emphasis has completely changed to getting more and more focused on power.”

However, there are technology solutions available, such as those mentioned above along with others such as discrete clock and voltage scaling, which Waran said are starting to find their way into the automotive domain.

In addition, cooling technologies are improving, and novel technologies like carbon nanotubes are being researched to address some of the most challenging issues. “While there are problems, the industry will find a way to solve them. If you look at some of the technologies that are coming out, like advanced packaging, those are all trying to address the same problem. It’s fundamentally a physics problem where there is a lot of die area, it’s dissipating a lot of heat. How do you pull that out? It’s the law of thermodynamics. I have a lot of heat generated in a small amount of area. How do I pull this out effectively? That’s where chiplets and things of that nature are coming into play. EVs are here to stay, and the progress is not going to be stunted because of these problems.”

But what about new use cases? And what about software-defined vehicle? What do you do if you wake up and your car doesn’t boot? These are issues that still need to be addressed, and they will need to be addressed across the automotive ecosystem.

“In the past, most of the architectures were defined at the OEM level or the tier one,” said Infineon’s Chettuvetty. “Then it would go to semiconductor companies, as a tier two. Those boundaries are getting diluted now, and there is direct interaction between OEMs and the semiconductor companies, which is a change. There needs to be much more collaboration so the OEMs know the capabilities of the semiconductor companies in the first place before they come with this architecture.”