Cars are getting smarter, even autonomous, thanks to artificial intelligence (AI), but those same technologies are being applied to modern vehicle design to improve workflows and create a circle of learning.
Alchip Technologies’ recently announced automotive application-specific integrated circuit (ASIC) platform is aimed at helping to streamline the design needs of automotive IC module and component manufacturers, as well as automotive companies. In an interview with Fierce Electronics, company VP Aston Tseng said the platform allows customers to design ASICs that meet all the necessary requirements of an automotive environment and ISO 26262, a functional safety standard, as well as other reliability demands.
The Alchip ASIC platform supports six key areas of automotive design, including autonomous driving and advanced driver assistance System (ADAS), safety, testing reliability, chip sign-off, and chip manufacturing. “We integrate a lot of knowledge which is used to do the design for automotive ASICs,” Tseng said. A critical piece of learning enabled by the platform is how an ASIC ages in the vehicle as times goes by, he said.
As vehicles get smarter, they become a reservoir of data that AI can learn from, Tseng said. “AI is good at data mining.” It can also link disparate sources of data together to get a more holistic understanding of what’s happening in the vehicles at a scale that human beings can’t. He said the ASIC itself has sensors to do detection that supports the functional safety requirements of ISO 26262.
The more data available, the more it’s possible to understand which step in the design or manufacturing process can be tweaked to make improvements, especially is areas of reliability and longevity, which are highly valued in automotive.
Tseng said a key trend in modern automotive design is the consolidation of computing architectures within the vehicle and the transmission of data to cloud for analysis, which supports continuous improvement.
AI tools span the automotive design lifecycle
There is no shortage of AI-powered tools for chip design, Tseng noted. Companies like Synopsys offer electronic design automation (EDA) tools that incorporate AI and machine learning to accelerate design process and workflows.
AI can enhance different kinds of automotive testing. Monolith was founded as a spin-off from a PhD project on uncertainty quantification and offers to help engineers to use AI to solve physics problems. One capability the company offers is New Test Recommender (NTR), which applies a proprietary active learning technology to help engineers find the best combination of test parameters to address key design considerations. Through iteration, models can be trained and improved to address unknowns – results are constantly re-evaluated to find the next set of parameters.
A specific automotive example for Monolith applies to electric vehicle (EV) batteries – battery performance could be evaluated across a set of dimensions to find the best charging profile to optimize the life of the battery over time, while also looking for protocols that could create unsafe thermal conditions.
One barrier facing the adoption of AI-enabled tools is standardization because standards for hardware in cars are still not completely sorted out and the compute architecture inside the car is still evolving to become more centralized. Pedro Lopez, director of automotive at RTI, told Fierce Electronics in an interview that this is slowing down development production capabilities. He’s active in organizations that are helping to solidify what a software-defined vehicle should be.
Tools need a roadmap
Because the standards inside the vehicle are flux, it’s difficult to settle on tools for the development phase and collaborate across the industry, Lopez said. Development using AI outside the vehicle doesn’t have a clear roadmap of the hardware interfaces, and everyone has their own toolbox. The next step is to standardize software interfaces, he said, so that it’s easy to create algorithms for test suites – right now, each automotive OEM has different APIs and data models.
Algorithms are available at a high level, Lopez said, “but we don't have a dictionary in automotive.” An AI dictionary for AI must encompass all automakers and include reusable models. If every vehicle maker and model has a different interface for testing, it’s going to be extremely costly, he said. “It's not going to make sense.”
AI is compelling in automotive design because the process is no longer linear, Lopez said. “Now you have this continuous integration, continuous deployment phase where you are developing, testing, releasing, testing, development, testing, releasing, and so on.”
Nvidia is supporting integration with its Omniverse computing platform, which enables individuals and teams to develop Universal Scene Description (OpenUSD)-based 3D workflows and applications. Danny Shapiro, VP of automotive at Nvidia, said everyone can plug in their 3D data no matter what tool they are using, so that everyone is putting data into a single master database. “If a change happens somewhere, it percolates throughout every single department,” he said.
The use of generative AI can be viewed as the natural evolution of computed aid design, Shapiro said, as it is becoming a co-pilot that helps automotive designers come up with ideas. Like ChatGPT, AI-enabled automotive tools need to be trained, he said, and that requires curated data so what they generate is reliable.
Generative AI acts as co-pilot
In an automotive workflow, what is generated might be inspired by a prompt to draw a rugged looking 4x4 or a sleek sports car, Shapiro said. If the AI has been trained on a huge body automobile knowledge, it understands diverse types of cars. “It understands the word rugged.” He said AI is valuable tool for concepting and brainstorming, and a wire frame sketch could be quickly transformed into a 3D or rendered model.
Shapiro said the AI could be trained on style of vehicle – if it’s trained on a specific carmaker’s body of work, and it can generate new ideas that align with their brand.
But generative AI is also revolutionary in that it can improve manufacturing, he added. “Factory planners can use generative AII to help lay out the factory in different ways to make it more efficient.”
Shapiro said there’s a clear need for better coordination between what design engineers are doing and integrating it with manufacturing. “You have these different silos within the automaker and these designers create stuff, but they're using different tools and different software applications.”