Automatic driving requires extremely powerful computing power. However, what remains to be discussed at this time remains: what processor architecture is best Sensor suited for running target identification and sensor fusion algorithms in order to implement computer-driven vehicles.

  Cadence has now introduced a processor based on a digital signal processor (DSP) architecture through its Tensilica. Tes Rieter's Vision C5 product is designed to run all the neural network layers of the Artificial Intelligence Engine.

  The most common approach to operating an artificial intelligence system is the use of a graphics processor (GPU) with the following advantages: The standard graphics controller Fuel Rail Pressure Sensor for conventional computers typically contains a number of parallel-running GPUs. After selecting the applicable algorithm, the graphics device can be reused by the artificial intelligence engine to provide superior performance advantages. At present, Nvidia (Nvidia) and other chip manufacturers have done very successful in this regard, but not everyone thinks this is the best way.

  Pulin Desai, director of product marketing for the Tensilica Vision DSP product line, commented that the method requires a very high-end GPU and is therefore very energy intensive. This approach may be applicable to high-end server farms where energy is not a critical factor. If it is applied fuel metering valve to the car, it is necessary to consider the embedded system space is limited, limited power and other issues.



  With Tensilica Vision C5 DSP, Kengeng used a very different way. C5 provides a high degree of centralized computing power - its size is less than 1 square millimeter, its performance has reached 1 TeraMAC (multiplication accumulation calculation step, Multiply-Accumulate computing steps), its high performance mainly due to its long instruction word (VLIW) vector processing instruction set (with 128-way, 8-bit or 64-way, 16-bit single instruction multiple data structure (SIMD) execution). The device Temperature Sensor is also optimized for applications requiring high visuality, radar, lidar and sensor fusion applications.

It is important to build architectures for multiprocessor designs - after all, the most advanced vehicle-mounted advanced driving assistance systems (ADAS), computer vision and other sensor information processing applications have increasingly relied on complex heterogeneous multiprocessing Device design. Compared to similar GPU solutions, C5 energy efficiency by nearly an order of magnitude. In the neural network hardware accelerator, the car also has a certain advantage: and professional, hardware-based accelerator is different Speed Sensor from the product can be reprogrammed, the design is more flexible and not out of date. At the same time, because of its functionality completely in the software to achieve, the company also uses a widely used development tools, and thus its development process is also easier. The final version of the C5 after the listing, the company will also provide neural network function library. The automotive application will be the largest market for this product. In addition, Keng Teng also plans to sell the product to manufacturers of UAVs and safety systems Throttle Position Sensor or other applications using neural network identification algorithms to perform target detection and target identification The