Teraki software eases hardware demands for autonomous driving
Berlin-based tech startup Teraki is tackling the challenge of exploding data analytics demands of the automotive industry by launching its breakthrough AI (artificial intelligence) and edge-processing technology on Infineon’s AURIX line of automotive microcontrollers. The startup’s Intelligent Signal Processing software is said to deliver a more than 10-times increase in efficiency of existing automotive chip, communications, and learning performance, making highly accurate AI applications possible in embedded environments.
The improvements engineered by the company could benefit a significant portion of the global automotive electronics market, which was estimated by Global Market Insights in September 2017 to reach $395 billion by 2024, and the predicted (by Navigant Research) 70% of cars sold in 2025 expected to be connected. The data processing challenge of self-driving cars is illustrated by the fact it will generate 60,000X more data than the average smartphone today. Geert-Jan van Nunen, Chief Commercial Officer of Teraki, said that a fully autonomous vehicle would produce 120,000 GB of data per month compared to just 2 for the average smartphone.
The aim of artificial intelligence is to create more intelligent machines by applying machine learning to enable vehicle customization and provide better speed and accuracy in embedded environments. Edge processing is a method of optimizing cloud-computing systems by performing data processing at the edge of the network, near the source of the data, said Daniel Richart, cofounder and CEO of Teraki.
“This reduces the communications bandwidth needed between sensors and the central data center/cloud by performing analytics and knowledge generation at or near the source of the data,” he added. “In the automotive example, many decisions will need to be made in the car. Both from an efficiency perspective: too expensive to do all computing in cloud. As from a performance perspective, it takes too long (or cloud communication may not be there) for use cases involving safety.”
The math behind the Teraki technology comes from cofounders Daniel Richart and Markus Kopf, and a team of more than 10 researchers. Richart comes from the Max Planck Institute of Quantum Optics in Munich working under Nobel Prize-winning atomic physicist Theodor W. Hänsch. Richart led research projects in quantum computing, a new field challenged by analyzing enormous volumes of data representing the multiple possible simultaneous combinations of quantum states of a particle.
Teraki is taking the technology honed at the high end of data analytics accuracy requirements and is scaling it for the highly constrained automotive infrastructure and, over time, other data-intensive IoT markets. It is aimed at OEMs and insurance providers that face an incredible opportunity to deliver innovative ways to use the vast amount of data generated by in-vehicle sensors, electronic control units (ECUs), and AI to improve vehicle safety and lower operational costs.
“We make predictive maintenance of the engine and crash detection possible using the existing hardware in the car,” said Richart, of two examples having the most potential. “We also help companies reduce bandwidth so the video transmission is real-time and also works in areas where there is less available network coverage. We have many more use cases.”
The historical challenge is that the high cost of expensive AI chips and the high computing demands of neural networks are preventing the widespread scaling of automotive AI applications, claim company experts. In addition, the limited processing power of ECUs, the bandwidth constraints of the in-vehicle CAN (controller area network) bus, the data communication costs of car-to-cloud networks, and the time required to train AI and machine learning components have been significant barriers to developing and scaling new—and often real-time—applications.
“Teraki’s offering is truly disruptive in that its dual objectives are to enable AI and edge processing in the increasingly data-driven automotive industry,” said Gibb Witham, Senior Vice President at Paladin Capital. “What is unique is that Teraki is taking technology honed at the very highest end of data analytics accuracy requirements and scaling it efficiently for the highly-constrained automotive infrastructure and, over time, other data-intensive IoT markets.”
In short, it is disruptive because it removes the reliance on expensive, high performance (AI) chips.
“Neural networks are big and energy consuming, with no higher accuracy for most automotive use cases,” said Richart. “So AI and neural networks are impractical and not scalable for automotive apps.”
Teraki is using its breakthrough edge-processing technology to downscale the cloud-analytics model to fit and operate with resource- and cost-constrained automotive ECUs and networks. The result is more than a 4-10 times increase in edge-processing solutions for automotive chip and data communications and more than a 10 times faster AI or machine-learning time performance.
”The biggest challenge for automotive system designers when implementing AI-driven applications is to find the balance between growing amounts of sensor data and the constraints of communication and processing technology,” said Ritesh Tyagi, head of the Infineon Silicon Valley Automotive Innovation Center. “Utilizing Infineon’s AURIX microcontrollers that support ASIL-D systems, Teraki delivers an innovative approach that significantly improves data analytics and enables true low-latency mobility services.”
The combination of these technologies translates into greater accuracy in detecting and responding to real-time events, resulting in higher levels of system reliability for applications such as accident detection, driver behavior identification, and predictive maintenance, added Tyagi.
Teraki has completed several preproduction validations by unnamed premium automotive manufacturers and their chip suppliers, as well as having many ongoing proofs of concept with additional OEMs.
Teraki announced in late September that it had raised $3 million in cumulative seed financing and government grants, with new investors Paladin Capital Group and GPS Ventures GmbH joining previous investors including Deutsche Telekom hub:raum. It will use this investment to support the growth of its automotive customer implementations and accelerate the development of its product roadmap.
“Our Intelligent Signal Processing software allows conventional sensors and ECUs to do far more; makes AI more practical, affordable and scalable; and significantly reduces CAN bus and car-to-cloud bandwidth constraints,” said Richart, summing up the company’s software value proposition.