- 99 per cent of current NNs execute within the aiWare3P core without any host CPU intervention
- The aiWare3P delivers NN acceleration for L2-L3 automotive AI-based systems
AImotive has introduced its latest release of aiWare3 NN (Neural Network) hardware engine. According to the company, this new design is incorporated with a range of new features. These features can optimise and improve the performance and power consumption of the device. The design features an enhanced host CPU offload combined with a simpler layout for larger chip designs.
Based on multi-core and multichip design
Each aiWare3P hardware IP core is said to offer up to 16 TMAC per second(>32 TOPS) at 2GHz. The hardware is implemented on a multi-core and multi-chip design. It is capable of delivering up to 50+ TMAC per second (>100 INT8 TOPS). Hence, it makes it ideal for multi-camera or heterogeneous sensor-rich applications.
The core of the hardware is designed according to the AEC-Q100 standards. This core makes the hardware resistive towards extended temperature operation. It also includes a range of features that enable users to achieve ASIL-B and above level certifications.
Extends the range of supporting hardware
It produces higher efficiency for a wide range of neural network functions. This is due to the improved on-chip data reuse and implementation of more sophisticated scheduling algorithms. The company has also upgraded its external memory bandwidth management. By this upgrade, it can now support a larger portfolio of pre-optimised embedded activation and pooling functions.
Runs on a real-time data compression method
It supports real-time data compression. This method reduces the requirements of external memory bandwidth for large networks. It creates cross-coupling between C-LAM convolution engines and F-LAM function engines. This, as a result, increases the execution efficiency of the system.
Based on the logical tile-based data management system
aiWare3P is based on a physical tile-based microarchitecture. It enables the physical implementation of large aiWare cores. It does so by minimizing the timing constraints on any process node. The hardware features logical tile-based data management. It enables efficient workload scalability up to a maximum of 16 TMAC/s per core. It eliminates the need for caches, NOCs or other complex multi-core processors-based approaches.
Marton Feher, senior vice president of hardware engineering for AImotive recently said in an interview,
“Our production-ready aiWare3P release brings together everything we know about accelerating neural networks for vision-based automotive AI inference applications. We now have one of the automotive industry’s most efficient and compelling NN acceleration solutions for volume production L2/L2+/L3 AI. When complemented by AImotive’s significant algorithmic, safety and production expertise for automated driving, we believe we offer our customers the most technology-rich automotive-focused solutions available today”.
Targets heterogeneous sensor applications
The aiWare3P hardware IP is being deployed in a range of L2/L2+ production solutions. These softwares are also being adopted for studies of more advanced heterogeneous sensor applications.
The company also informed that their upcoming products will devices which can bear high-resolution inputs in order to exceed public benchmarks. The aiWare3P RTL will be in stock for all customers by January 2020.