Advanced Packaging For Automotive Chips

Multiple types of chips may be better than one for dealing with large amounts and different types of data, but in automotive applications it’s not entirely clear how or even whether they should be packaged together.

The biggest problem with electronics in vehicles is the extreme range of temperatures, both within and outside of vehicles. Without adequate cooling, chips can age prematurely, signals will drift, and in some cases the chips may cease to function properly. Still, cooling adds complexity, cost, and another possible point of failure, making this a difficult decision for carmakers and chip designers.

“No matter where you look, every module has thermal constraints, because every module is enclosed,” said Willard Tu, senior director in the Automotive Business Unit at Xilinx. “They’re always trying to jam them into weird little spots. Some modules have a disadvantage in that they’re more subjected to ambient heat build-up.”

For example, the forward-looking camera on the windshield is in a particularly harsh environment, because the sun beats down on the windshield. “This means the module is naturally heating up by itself, not just from the thermal of the electronics, but from the environmental conditions,” Tu said. “Whether it is hot or cold, temperature swings also create challenges.”

Proximity to the engine makes package design especially challenging. “The engine gets hot, and there’s power circuitry,” said Marc Swinnen, product marketing director at Ansys. “At the same time, there are a lot of electronics in the car that are not that close to the engine, such as the infotainment system. Also, the self-driving electronics are not necessarily intimately tied to the engine, so the special thermal constraints only apply to a subset of the electronics. The rest can be housed in a normal electronic housing, under the seats or in the dashboard.”

While these kinds of decisions may not even hit the radar of most design teams working in consumer electronics, the extended lifetime of cars and trucks and the amount of data being processed in vehicles requires some sophisticated design and cooling techniques. The problem is there is little history for advanced chip designs under these kinds of extreme conditions, and there are no designs in use today at the most advanced process nodes or in advanced packages.

Consider chips that include a high degree of signal processing, such as 4D imaging radar or lidar.

“Here, thermal is always an issue,” said Tu. “Traditionally, the industry solved that by going down Moore’s Law to a smaller transistor gate length. There, the active power is always lower when the device is running. The static power, when the device is just sitting doing nothing, is actually a little bit higher, but the device is usually ‘on’ most of the time, not ‘off.’ We only worry about ‘off’ when it’s an application that might impact the battery. This is an area of interest because most carmakers are increasing the size of batteries.”

Moore’s Law historically has provided a side benefit of reduced heat and better performance, but the benefits are diminishing at each new node. “Not everybody’s moving down that lithography node from 28nm to 16nm to 7nm to 5nm to 3nm, and so on,” said Tu. “The challenges now are becoming cost-oriented, which means a lot of companies are shying away from moving to the 5nm or 3nm node because it may not be affordable for the market any longer. As a result, they are looking to deal with thermal mitigation in other ways, and are beginning to look to the system level instead of shrinking.”

This plays to the strength of FPGAs, which can perform more compute per clock cycle than traditional ASICs. “When you look at thermal dissipation, one of the biggest contributors is the clock frequency of the SoC,” Tu said. “If the device is operating at 1 or 2 GHz, that’s the main generator of the thermal computation power. But in an FPGA, instead of having one pipe coming through, it can be configured to have five or six pipes. So for every clock cycle, five or six instructions are performed. Instead of having a CPU or multiple CPUs running on one clock frequency, the device can be figured to have 10 or 20 specialized pipelines that are doing signal processing very heavily.”

It also plays to the strengths of advanced packaging in some parts of a vehicle. But some approaches will work better in certain parts of a vehicle than others, and this is only now being worked out.

Strategies for addressing thermal
Thermal challenges are clearly at the top of the list for all chipmakers targeting the automotive market, as well as for companies developing advanced package technology.

“The basic challenges are always going to be the same for every sector where it is used,” said Melika Roshandell, product marketing director for multi-physics system analysis at Cadence. “In automotive, the challenges come in because you start at a higher ambient temperature, which means the challenges would be a bit higher. But whatever strategies other industries are using can apply to automotive. Automotive also has the benefit of a bigger space, so other thermal improvements can be used in the design more than, say, a mobile phone could. It means a fan is an option. Heat sink? That’s an option, as well.”

Making this work as planned requires documentation of the thermal on different electrical parts. In the past, this was done using a steady state analysis of the PCB, but that does not provide enough detail to make that assessment.

“For this reason, electro-thermal co-simulation is becoming very important in automotive because of the reliability requirements,” Roshandell said. “You don’t want to throw out your car in two years. You want it to last for a long time. This type of co-simulation shows how over a certain amount of time, in a transient way, the temperature impacts heating and IR drop, etc.”

Fig. 1: Different advanced packaging options. Source: Cadence

Even though the automotive sector has little experience with advanced electronics, this process is well understood. “The good thing about the automotive industry is there’s an awful lot of regulation, which forces a lot of analysis and a lot of bookkeeping,” said John Ferguson, director of product management at Siemens Digital Industries Software. “They have pretty well characterized the temperature conditions and what will cause those temperature conditions within their system. Once you know that, then you’re at least one step closer to asking how to design a stack of chips that could go into that situation.”

Heat and memory
One of the key concerns with advanced electronics is the effect of heat on memory. While DRAM is considered extremely reliable, it becomes less so in extremely hot environments. This is particularly troublesome in automotive applications because an enormous amount of safety-related data generated from sensors throughout a vehicle needs to be processed and stored somewhere, and access needs to be extremely rapid. In addition, that data needs to remain intact and uncorrupted.

One possible solution involves high-bandwidth memory (HBM), which is basically DRAM chips stacked inside a module. “Users have been considering putting HBM into vehicles since HBM2,” said Brett Murdock, product marketing manager for memory interface IP at Synopsys. “The real challenge is on the memory side because heat in the automotive environment makes it challenging. It gets very hot very quickly, and memories stink at being hot. Interestingly, with HBM3, it’s been shown in the standard task group that there is a lot of interest to work out the automotive temperature ranges for HBM3 devices. If the vendors can solve the automotive problem on their side, that can open a high demand market for HBM devices.”

For this to happen, they must be able to withstand automotive-grade temperature swings. “Today, the number one use of HBM is in graphics processors, complete with massive heatsinks and/or liquid cooling,” said Murdock. “That’s the number one problem to solve. HBM initially was supposed to be a 3D technology, meant to sit right on top the CPU, but that wasn’t possible because of temperature. It got too hot, and that’s why it went to 2.5D using an interposer. You’re dropping a massive heatsink on top of the overall package to bleed all that off, which means the stacked die is a problem. Then putting that in an automotive environment is going to be a problem. Vendors are trying to approach the issue because they see the demand there, and they see the opportunity.”

Using an interposer adds another potential problem from a reliability standpoint. “The mechanical connections for the automotive environment for the interposer are something that you deal with when you’re doing boards, but I don’t think anybody has really gone all through the thought process or the qualification process for what the interposer will look like, and if it’s okay,” he said. “But this is a much smaller problem. If the heat problem is solved, the mechanical problem is secondary.”

If these problems can be solved, the ecosystem could very quickly create automotive memory IP for HBM3. “In late 2023/early 2024 timeframe we might actually start to see automotive design starts with HBM with qualified automotive parts,” Murdock said.

Stacking die
So far, true 3D packaging in automotive chips isn’t happening, and it’s not obvious it ever will because of issues such as thermal dissipation and form factor.

“It’s probably easier to have very thin components that you can put almost anywhere within the engine or within the car itself,” said Siemens’ Ferguson. “If you’ve got something that’s a half an inch, or an inch tall, it might be more difficult to squeeze in. That certainly would be one of the considerations.”

But the vehicle is only one part of the automotive infrastructure. Stacked die could play a significant role inside of data center, which will collect and quickly process data generated by millions of vehicles.

“If you think of a carmaker like Tesla, there’s the electronics in the car, which have to be highly reliable and immune to vibration and heat, be somewhat lightweight, and not require a lot of cooling,” said John Park, product management director for IC packaging and cross-platform solutions at Cadence. “But that car talks to a server farm, where there’s a bunch of AI stuff happening. An image is sent quickly to this AI farm, which can be in a liquid-cooled type of environment. In the car itself, I don’t see a big reason to go to a lot of 3D integration, not just because of thermal but because of cost, reliability, and other things. However, at the home hub that a Tesla is talking to, that’s a different animal. If you’re an automotive company, I don’t think that’s something you’re going to be sticking in your mainstream cars. In the case of a company like Tesla, and all of the autonomous driving/autopilot things that people want to do, people are willing to pay a premium for that. There, it probably makes some sense for those types of cars that are being built.”

More likely, different multi-die approaches will be used inside of vehicles. “There’s definitely an evolution here,” said Brad Griffin, product management director for multi-physics system analysis at Cadence. “Maybe the need to do the heterogeneous integration in automotive has not been as demanding as in smaller devices, but we are seeing more components in the automotive market becoming electric such that there is an evolution toward at least 2.5D integration with silicon interposer.”

Work is underway in the automotive ecosystem on ECUs and other types of devices that are going to be used in automotive applications. “A big part of what is happening in that migration is the concern about being able to electrically model how everything will work,” Griffin said. “Engineering teams are used to having a bit more space between the signals going between the different die or components, and with everything so compressed in a 2.5D, 3D IC, or mixed die stack, there are a lot of signals that go between those. They’re very close together on a silicon interposer, so the modeling and extraction of those signals has to happen. And not just like one or two, or the worst case. They want to do the full data bus or the full system itself.”

Cosmic effects
There are other problems to resolve, as well, and this is an area where different packaging technologies play an especially important role. As more electronics are added into vehicles, they are subject to the same kinds of cosmic effects as other electronics. The more advanced the chips, the denser the circuits, and the more likely that a single alpha particle can flip multiple memory bits rather than just one.

“These are shooting through us all the time, and if they hit the right memory bit at the right time, they can flip that memory bit,” said Ansys’ Swinnen. “So any flip flop in the chip can, at any time, randomly flip if it gets hit by the right cosmic ray at the right time. In mil/aero applications, they try and protect devices by putting the circuitry in metal boxes that can afford some level of protection from these particles.”

Designers of space electronics are very aware of this effect because of the high radiation in the upper levels of the atmosphere and in deep space. “Here on Earth, it’s an IT transient failure mechanism that you can solve by shielding. That’s typically not done, so you just live with it,” Swinnen said. “But it could happen anytime. This is one of the reasons why, when the original space shuttle went up, there were five redundant computers. Four were identical and the fifth was different. The reason was exactly that, because you’re up in that high radiation zone. You could get a wrong answer from any of the four, at any time. The fifth was the tiebreaker.”

This is a big concern in automotive today, as well, but it’s also an area where advanced packaging may help. “This is worse in cars in the sense that you’re less tolerant,” he said. “If your phone glitches, and does something weird, oh well, you just call back. But if your car glitches, it might be a lot worse.”