Provide the fastest response and the most energy efficient neural network acceleration function for the next generation of automotive application scenarios

Imaginationtechnologies announced the launch of a new generation of neural network accelerator (NNA) product imgseries4 for advanced driving assistance system (ADAS) and automatic driving applications. Series4 can provide powerful support for leading automotive industry subversives, first tier suppliers, OEM and SOC manufacturers.

Series 4 has a new multi-core architecture, which can provide 600tops (trillions operations per second) or even higher ultra-high performance, and can provide low bandwidth and extremely low latency for large neural network workload.

The automotive industry is in the teeth of the storm of change. New applications such as autopilot and autopilot have raised higher requirements for the performance level of AI. To this end, imagination has partnered with leading companies and innovators in the automotive industry and other industries that focus on functional safety.

Series4 has started to offer licensing and will be available on the market in December 2020.

Imagination’s low-power NNA architecture aims to perform complete network reasoning while meeting the functional security requirements. It can perform multiple operations at once, maximizing per watt performance and providing industry-leading energy efficiency.

Series 4 has the following features:

  • Multi core scalability and flexibility: multi core architecture supports flexible workload allocation and synchronization among multiple cores. Imagination’s software provides fine-grained control, and improves flexibility by batching, splitting, and scheduling multiple workloads. It can now be used on any number of cores. Series 4 can configure 2, 4, 6 or 8 cores for each cluster.
  • Ultra high performance: each single core of series 4 can provide 12.5tops performance with less than one watt power consumption. For example, an 8-core cluster can provide 100tops of computing power, so a solution with 6 8-core clusters can provide 600tops of computing power. In AI reasoning, the performance of series 4nna is 20 times faster than that of embedded GPU and 1000 times faster than that of embedded CPU.
  • Ultra low latency: by grouping multiple single cores into 2-core, 4-core, 6-core or 8-core multi-core clusters, all cores can cooperate with each other to process a task in parallel, reduce processing latency and shorten response time. For example, for an 8-core cluster, the latency would ideally be reduced to 1 / 8 of that of a single core.
  • Save a lot of bandwidth: imaging’s tensortiling (ITT) is a patent pending technology, which is also a new function in series 4. It can tiling computing tasks, make full use of on-chip storage, improve data processing efficiency, and save bandwidth for accessing external storage. ITT uses the dependence of local data to store intermediate data in on-chip memory, which can minimize the transmission of data to external memory, thus reducing the bandwidth by up to 90%. ITT is a scalable algorithm, which has a significant advantage in the network with a large number of input data.
  • Vehicle specification level security: Series 4 includes IP level security functions, and the design process conforms to iso26262 standard, which can help customers obtain iso26262 certification. Iso26262 is an industry safety standard to solve the risk of automotive electronic products. Series 4 can safely perform neural network reasoning without affecting the performance. Hardware security mechanism can protect compiled network, network execution and data processing pipeline.

“Although we still expect the demand for ADAS to increase by two times by 2027, the auto industry has already turned its attention to the further automatic driving cars and autopilot taxis,” said JamesHodgson, ABIResearch intelligent travel and chief automotive analyst. In the process of evolution from L2 and L3 ADAS to L4 and L5 automatic driving, the wide application of neural network will be a crucial factor. These systems will deal with hundreds of complex scenes, extract data from a large number of sensors such as multiple cameras and lidar, so as to achieve automatic valet parking, intersection management and complex urban environment security navigation solutions. The combination of high performance, low delay and high energy efficiency will be the key to the realization of highly automatic driving. “

AndrewGrant, senior director of ImaginationTechnologies AI business, said: “we believe Series4NNA will become an industry standard platform for the development of advanced driver assistance systems and autonomous driving vehicles. Some innovators are already building chips that support the next generation of ADAS and self driving cars. It’s time for any company or R & D team that wants to play an important role in the automotive industry to integrate this technology into their platform. “

For more information about series4nna, please click the link below to view our theme conference.

Leave a Reply

Your email address will not be published. Required fields are marked *