With the automated driving industry changing pace, the semiconductor industry must also adapt, to bring the right hardware solutions to challenging conditions.
Written by László Kishonti from AImotive.
Back in March, I wrote about the importance of collaboration for survival in the automated driving industry. A lot of what I spoke about then is becoming a reality. Over the summer, new cooperations and acquisitions have been announced almost weekly. Just recently week Toyota announced it will work with Suzuki to bring automated driving tech to consumers.
Changing front lines — The challenge remains
What I didn’t address then was how the semiconductor industry will have to react to the changing landscape. However, this question was placed front and center by Tesla’s announcement of the FSD. (Jump back to my previous blog to learn more about how the FSD relates to aiWare and our research into neural network acceleration.)
A year ago, many considered hardware performance one of the largest bottlenecks of fully self-driving technology. This may lead you to believe that with the focus shifting from autonomous vehicles to gradually increasing driving automation the challenges of hardware platforms can be pushed aside.
While great for quick and effective prototyping, a setup like this one would be a limitation to the adoption of automated driving systems. Optimized hardware could offer better performance in a smaller form factor with better thermals.
However, this is not the case. To continuously increase the level of automation in road vehicles, hardware platforms have to keep up. Moreover, they have to have performance readily available to support new features delivered as over the air (OTA) updates. To do this they need scalable, flexible, upgradeable purpose-built platforms to accelerate NNs. The catch is, most OEMs simply don’t have the expertise, or time, to get this done in-house.
This is why even the biggest names are turning to outside parties for support. Remember that Toyota also announced it was working with DENSO to create advanced chips for automated driving.
Collaboration — A friendly battering ram
I’m sure we’ll see several other such ventures pop up in the coming months, years as advancements in automated driving functionalities bring the need for increasingly more advanced hardware to the fore. But why is cooperation the solution? Three benefits come to mind:
- Bringing the best minds in various fields of expertise together to find the best solution to a problem;
- Sharing the financial and commercial risks and costs while equally benefiting from joint success;
- Creating standardized, or semi-standardized solutions through broader industry collaboration.
The right minds for the right challenges
Automated driving is possibly one of the most demanding use cases for hardware in the automotive industry. That means the most demanding use case in a fundamentally demanding industry.
First, the general safety and redundancy requirements of the automotive industry are to be considered. Hardware platforms must be ASIL-B to ASIL-D compliant, with fallbacks and redundancies in place to ensure continuous operation.
Second, they have to be built to withstand the bumps and potholes of road networks within extremely demanding thermal limits. And they have to do all this while offering the highest possible level of performance in the smallest possible form factor, with the lowest possible power consumption.
Third, they have to get neural network acceleration right to ensure minimal latency while computing some of the most advanced artificial intelligence algorithms. When rolling down a highway at 70 mph, whether information is processed within a few milliseconds or after nearly a second makes a huge difference.
And then come the challenges listed above: scalability, upgradeability, and flexibility.
Consider how many different fields of expertise are required to get every aspect of this problem right. How many companies in the world have experience in each of these? How many have the resources to single-handedly build teams and fund research projects to solve them? With inter-industry collaboration, the automotive, semiconductor and automated driving industries can find the best solutions to the right problems.
The right minds for the right challenges
Automated driving is possibly one of the most demanding use cases for hardware in the automotive industry. That means the most demanding use case in a fundamentally demanding industry.
First, the general safety and redundancy requirements of the automotive industry are to be considered. Hardware platforms must be ASIL-B to ASIL-D compliant, with fallbacks and redundancies in place to ensure continuous operation.
Second, they have to be built to withstand the bumps and potholes of road networks within extremely demanding thermal limits. And they have to do all this while offering the highest possible level of performance in the smallest possible form factor, with the lowest possible power consumption.
Third, they have to get neural network acceleration right to ensure minimal latency while computing some of the most advanced artificial intelligence algorithms. When rolling down a highway at 70 mph, whether information is processed within a few milliseconds or after nearly a second makes a huge difference.
And then come the challenges listed above: scalability, upgradeability, and flexibility.
Consider how many different fields of expertise are required to get every aspect of this problem right. How many companies in the world have experience in each of these? How many have the resources to single-handedly build teams and fund research projects to solve them? With inter-industry collaboration, the automotive, semiconductor and automated driving industries can find the best solutions to the right problems.
Whether OEMs follow a centralized or distributed approach to processing hardware, they will need scalable and flexible hardware platforms however they are deployed in vehicles.
Minimal risk for mutual benefits
Such a concentration of expertise would also mean that cooperation between different companies would have a higher chance of creating a scalable, efficient and affordable solution to this challenge. Furthermore, without having to build their own teams from scratch, and sharing the R&D investments, companies will have to invest less for a higher likelihood of positive results.
Of course, one can claim that a proprietary solution offers better differentiation and, as a result, a higher profit margin but the risks are also much higher. The costs of developing such a chip are in the magnitude of hundreds of millions of dollars. With so many unknowns and moving variables in the future of driving automation, such an investment is almost a double or nothing bet. Not to mention, it is the capabilities of the automated driving software stack that will provide the most area for differentiation.
Sharing is caring
The automotive industry has a long-standing tradition of standardization. The enormous network of standards benefits OEMs and consumers alike. They guarantee increased safety, the international operation of vehicles, and increase transparency and accountability. A wide cooperation of stakeholders could create a standard or set of standards for the new challenges facing automotive hardware. Such standards, for example, could define interfaces to ensure the upgradeability of platforms or performance overheads for the safe deployment of over the air updates. Beyond increasing safety, such standardization would also reduce the need for proprietary research in various areas.
Getting automotive hardware right
The considerations above are why we at AImotive are proponents of an automated driving chip consortium. We are uniquely placed to be part of a wider collaboration in the field through our expertise in both developing automated driving software, and a neural network accelerator hardware IP core for automotive use cases. Furthermore, relying on our aiSim simulator we have the ability to conduct hardware- and software-in-the-loop tests of developing hardware solutions to benchmark their performance against real-world automated driving workloads.
Hardware platforms for automated driving pose a unique challenge that requires experts from several different fields to come together. To ensure safety, upgradeability, time of life, scalability, and flexibility we believe it is best for the industry to work together. This is the only way to get automotive hardware truly right.
About AImotive:
AImotive is one of the largest independent teams in the world working on automated driving technologies. Developing self-driving software, proprietary simulation tools and neural network acceleration hardware IP, the company is building an ecosystem to aid the deployment of automated driving. To catalyze industry collaboration AImotive’s products are scalable, modular and hardware agnostic. The company currently has over 220 employees, working at offices in in Mountain View, California; and Yokohama, Japan. The bulk of development happens at AImotive’s headquarters in Budapest. Among them are over thirty specialized artificial intelligence researchers, while 20+ members of the team hold PhDs.