Latency, the fractional delay between data capture and executing a command, holds significant importance in SAE Level 2 and 3 advanced driver-assistance systems (ADAS). Shaving milliseconds off a vehicle’s reaction time to a hazard could be the difference between life and death. However, as processing information from multiple sensors becomes more energy intensive, some developers are moving machine learning (ML) algorithms away from the edge.
Harald Kroeger, President of Automotive at SiMa.ai, tells Automotive World that this is a mistake. “Relocating the vehicle’s brain far away creates latency, which is very harmful in ADAS. You could crash into a tree, and the car would only know it 30 seconds later.” As such, his company has developed an ML system-on-chip (MLSoC) with accompanying software to keep processing in vehicles and “close to the action” instead of siloed in data centres.
Installing a ‘guardian angel’
The artificial intelligence (AI) models used for automated driving can be large and power hungry, which not only lowers range when installed in electric vehicles but also slows down inferencing. Kroeger states anything in excess of 100 milliseconds delay can lead to undesirable outcomes in most driving situations, and the degree of ADAS latency is contextually determined by factors like environment, vehicle speed, and frames per second (FPS) performance limitations.
Addressing the latter has become a focus point for many tech companies. “There’s even an industry ‘Olympics’, the MLPerf Benchmark, to compete on how many FPS can be run on a certain model,” he says. At the time of writing, SiMa.ai has won three times in the AI/ML edge inferencing category. In March 2024, it achieved 150 FPS per watt, besting comparable offerings from Dell and Qualcomm by 200-300%. The safety implications of this performance improvement are significant.
In 2023, the World Health Organisation recorded 1.2 million fatalities resulting from road traffic collisions. Some analysts question whether the current generation of partial automation systems have any definite benefits on road safety. Robotaxi operator Waymo published a study in December 2024 claiming Level 4 autonomous vehicles (AVs) can be safer than human drivers, but the rollout of this technology has been slow, limited, and far from incident free.
Kroeger believes widely deployable, high-performance, AI-driven edge inferencing is the essential component missing from ADAS and AV systems. “We need to install a guardian angel inside every vehicle. If routine instances of driver distraction and error can be corrected, we could save so many lives.”
Performance, efficiency, applicability
SiMa.ai is positioning its unique MLSoC platform as the gateway for unlocking next-generation ADAS/AV performance. Importantly, it has been conceived specifically for automotive applications. “A lot of the technology out there was designed for something else and retrofitted,” says Kroeger. Comparatively, SiMa.ai’s ‘blank slate’ enabled it to create a solution from the ground up, unlocking both performance and cost efficiencies. “Affordability leads to adoption; if something’s too expensive, no one will use it.”
SiMa.ai’s MLSoC has an ML performance of 50 TOPS and a ResNet-50 (an image classifying architecture) performance of more than 300 FPS per watt. The company claims latency can be reduced by a factor of ten, with all compute in Level 2/3 systems achieved at less than 25W, or less than 100W at Level 4.
Equally important, Kroeger adds, is applicability. “We recognised early on that focusing on one specific model is not a good idea. Doing so means that every new development makes the hardware redundant.” SiMa.ai designed the chip according to three foundation pillars: to be compatible with any computer vision and generative AI application using ML at the edge, regardless of vehicle type; to offer best-in-class performance per watt; and to provide an interface usable by anyone instead of a small demographic of hardware experts.
The secret was developing a chip that can be adapted quickly and easily through software—SiMa.ai considers itself primarily a software company that also builds silicon products. Offering a general compute platform that’s highly malleable according to use case future-proofs its solution in the rapidly evolving vehicle AI market. “Putting a brain in cars will fundamentally change how customers interact with them on a daily basis,” states Kroeger. “The industry talks about software-defined vehicles (SDVs), but AI-defined vehicles represent a far larger shift.”
The AI-defined vehicle
Although SDVs currently lack a coherent definition within automotive, Kroeger states that OEMs and suppliers are enthusiastic about SiMa.ai’s MLSoC and the possibilities of an AI-defined vehicle. “The reason is that pretty much everyone is unhappy with the power consumption, restricted versatility, and lack of focus on automotive that current solutions provide.” Now that SiMa.ai has gained visibility in a highly competitive space, it is pushing to take on the Big Tech players.
Everyone now realises that software, not hardware, will make the magic happen
In December 2024, SiMa.ai combined its AI/ML capabilities with silicon design and verification firm Synopsys to maximise customisation for automotive-centric IP, subsystems, chiplets, and SoCs. “Synopsys is almost like a superpower; there’s pretty much no chip in the world in which it wasn’t involved.” Notably, the company’s advantage in virtualisation will enable simulated chip functionality testing, resulting in faster and better software iteration. This type of development process is an emerging trend within SDV innovation.
Kroeger predicts that ADAS systems integrating the SiMa.ai MLSoC could be on the market by 2029, with the potential for introducing more limited AI functions into cars even sooner. “A breakthrough in embedded AI can only happen when automakers reconsider performance and energy efficiency,” concludes Kroeger. “Future-proofing vehicle architectures for the big changes coming is vital, and everyone now realises that software, not hardware, will make the magic happen.”
