So far, information from chip vendors such as Nvidia, Mobileye and NXP seems to show that their respective autonomous vehicle platform concepts (and how they are intended) are very different. This is understandable in view of the fact that everyone will use their existing and what they believe can defeat their opponents to seize market position. However, it is worth noting that for the original car manufacturers and first-tier auto parts suppliers, the challenge is the same: the number of electronic control units (ECUs) in the car is increasing, and there are various auto-driving vehicles. Sensors, the collected sensory data needs to be processed, analyzed and fused, as well as security issues - the door of the connected car. Those challenges are closely related to advanced visual processing techniques, deep learning, mapping, and more, and also affect the processor performance requirements of the new system architecture. Will this look like Google’s self-driving vehicle? So, there is a $64 million question here – today's automakers and first-tier auto parts suppliers already know about the 2020 autopilot vehicle architecture? When Eric Baissus, CEO of French startup IC design company Kalray, recently accepted an editorial interview with EE TImes, the answer to the above question was: they didn't know, or didn't know it; and that's why the startup thinks it's equipped with 288 A Massively Parallel Processor Array (MPPA) of the VLIW core has come to a good time to enter the market. Kalray originally developed the ultimate computing technology needed for nuclear bullet simulation for the French Atomic Energy Commission (CEA); the company now targets key embedded markets (such as aerospace) and cloud computing. Baissus believes that autonomous vehicles are also part of a key embedded market, as such vehicles need to absorb large amounts of data from outside the vehicle and in various parts of the vehicle, process them quickly, and then make quick decisions; The automotive industry: "The need to handle multi-domain integration (mulTI-domain funcTIon integraTIon) and a new generation of processors that can perform processing tasks at very high levels." Of course, the so-called "manycore revolution" has begun; but Baissus said: "No one has successfully designed a "supercomputer single chip" with massive parallelism and more than 100 cores;" Kalray's latest generation of 288 cores The processor Bostan integrates 16 processor clusters each with 17 cores, 2MB of shared memory (SMEM), data transfer speed of 80GB per second, and 16 system cores. In addition, Bostan is also a network single chip that can adapt to critical moments and supports high-speed Ethernet interfaces (8x1 GbE~10GbE). The chip is also equipped with high-speed encryption and decryption, as well as easy link function with GPU/FPGA accelerator. Therefore, the MPPA architecture can provide DSP type acceleration with power saving, timing predictability, and multi-domain support (for example, different processor clusters can implement different embedded systems used in different parts of the car) And scalable massively parallel operations (internal processors can be combined to accommodate the complexity of the system). This "supercomputer single chip" for autonomous vehicles is not very similar to Nvidia's Drive PX platform? Nvidia calls Drive PX "the world's most advanced self-driving vehicle platform", which claims to combine deep learning, sensors, surround video and more. In this regard, Baissis explained that there are two differences between the two: First, Kalray's solution is “certifiableâ€: “I mean we can prove determinism and guarantee timing. In high-performance computing, a one-second delay is acceptable, but in critical embedded markets—such as aerospace and automotive—only 10 millisecond delays can be fatal.†Secondly, he said that engineers need to understand CUDA if they want to use Nvidia chips, but: "Our chips can use standard tools and Linux to execute standard C/C++ code;" Automakers already have a lot of old code written in C and The algorithm, even if the car manufacturers turn to the new autonomous vehicle platform, the old program code is still very important. Not only does Nvida expect future cars to require more processing power, but another chipmaker, Mobileye, also recently "pre-released" the EyeQ5 processor and promised to offer chip engineering samples in 2018. The EyeQ5 is designed with an advanced 10nm or less FinFET process and will feature eight multi-threaded processor cores and 18 Mobileye next-generation vision processor cores; the company says the EyeQ5 can perform 12 Tera operations per second while Can control power consumption below 5W. And everyone including Baissus dare not to sneak into Mobileye; unlike Nvidia's Drive PX, which is seen by many industry observers as a "testing platform" for autonomous vehicles, Mobileye follows the commercial market's demand for higher processing performance. A trend toward lower power levels is also required; by leveraging proven visual processing algorithms, EyeQ5 combines data—incorporating 20 external processors such as cameras, radar, and opticals—in a single chip. But can EyeQ5 manage the ECU inside the self-driving vehicle? A Mobileye spokesperson explained to EE Times that EyeQ5 not only supports data fusion, but also performs decision-making, but the decision-making action is carried out in other aspects—that is, the low-end ECU chosen by the car manufacturer. Kalray's role in its many core processors is slightly different from Mobileye and Nvidia. Baissus said that there are already many advances in sensors and machine learning algorithms that are necessary for autonomous vehicles: "But there is no practical work in the processor field;" This is where he sees opportunities. Baissus believes that a new generation of autonomous vehicle processors need to perform beyond data fusion: "They must be more like an open platform;" and he expects to provide an open processing hub for autonomous vehicles - it can be called "Super ECU". This super ECU provides cross-domain integration on a single chip, delivering better results for key elements including sensing, learning, security, networking and cost. Without revealing the name of the manufacturer, Baissus said that the leading automakers and first-tier auto parts suppliers are using the Kalray platform to prototype the first autopilot, but he also admits that the current autonomous vehicle architecture is not mature enough; Through cooperation with a number of major manufacturers, he hopes to understand the needs of more car manufacturers to help the company define the next generation of autonomous vehicle solutions. Baissus does not rule out licensing the MPPA architecture to other automotive chip manufacturers. The mono-crystalline wafer production flow consists of cutting, cleaning and sorting procedures. The production capacity and wafer yield rate have been continuously improved to meet the clients` requirements on wafer quality and yield with great endeavors to constant improve our cutting technology and final inspection capacity. Solar Cell,Photovoltaic Cell,A Grade Solar Cell,High Power Solar Cell,Battery Bank Power Wuxi Sunket New Energy Technology Co.,Ltd , https://www.sunketsolar.com