In the field of artificial intelligence, the realization of the algorithm depends on the powerful running speed of the computer, so the chip is particularly important. At present, the AI ​​chip market has attracted many players. Both traditional semiconductor companies and so-called start-up companies have begun to defect to this next gold mine. This article provides an overview of the typical chip vendors and their products in the field of artificial intelligence.
1. Nvidia
With an image processor with recognition and marking capabilities, Nvidia takes control of this opportunity before AI has fully emerged. In 2016, Nvidia also released a number of chips for deep learning, such as a Tesla P100 GPU for performing deep learning neural network tasks released in April, and Tesla P4, a Pascal-based deep learning chip, released in September. TeslaP40, where Pascal architecture can accelerate deep learning by 65 times.
In addition to R&D chips, Nvidia also released multiple hardware and platforms for different areas, further expanding its artificial intelligence layout. At CES 2017, Nvidia released the XAVIERAICARSUPERCOMPUTER, a smart home hardware chip, and an onboard computer with the artificial intelligence system ProAI (developed jointly by ZF and Nvidia). It is reported that the ProAI system can process data from car sensors and cameras through deep learning, can clearly identify the surrounding environment and accurately locate on high-definition maps, and plan a safe road ahead for vehicles, which is further suitable for automated driving of expressways.
2, ARM
So far, 85% of the world's smart mobile devices have adopted the ARM architecture, of which more than 95% of smart phones use ARM processors. Today, ARM has an absolute position in the rapid development of intelligent hardware and the Internet of Things.
In addition, according to its 2015 Q4 financial report, ARM-licensed chips are mainly used in mobile computing, smart cars, security systems, and the Internet of Things. In the field of smart cars, including NVIDIA and Qualcomm, ARM-based supercomputers for driver assistance systems have been developed and designed. Earlier, for the acquisition of ARM, Sun Yat-sen, Softbank CEO has clearly stated that ARM chips will promote artificial intelligence to a singular point. After the acquisition, Softbank also expressed great support for ARM's long-started artificial intelligence project "BlueSkyProgram".
3, Intel & Nervana
In November 2016, Intel Corporation released an AI processor called Nervana, which they claimed will test the prototype in the middle of next year. If all goes well, the final form of Nervana chips will be available in 2017. The chip is based on a company Nervana previously purchased by Intel. According to Intel's people, this company is the first company on earth to create chips exclusively for AI.
Nervana has been trying hard to introduce the machine learning function into the chip. It is an artificial intelligence ASIC chip supplier. With support from Intel, Nervana is planning to launch its custom chip NervanaEngine for deep learning algorithms. According to Nervana related personnel, NervanaEngine can improve its training performance by 10 times compared to GPU.
4, IBM
The hundred-year-old giant IBM has released wtson long ago. Now his artificial intelligence machine has long invested a lot of research and development. Last year, he couldn't hold back and devoted himself to the research and development of human-like brain chips. That is TrueNorth. The stamps are only a few grams in size and weight, but they have integrated 5.4 billion silicon transistors, built-in 4096 kernels and 1 million. The "neurons", 256 million "synapses," are equivalent to a supercomputer and consume only 65 milliwatts of power.
TrueNorth is the latest result of IBM's participation in DARPA's research project SyNapse. This chip treats digital processors as neurons and uses memory as a synapse. Unlike the traditional von Neumann architecture, its memory, CPU, and communications components are fully integrated. Therefore, the processing of information is completely performed locally, and since the amount of data processed locally is not large, the bottleneck between the conventional computer memory and the CPU no longer exists. At the same time, neurons can communicate with each other conveniently and quickly. As long as they receive impulses (action potentials) from other neurons, these neurons will act at the same time.
5, Google
Google's artificial intelligence related chip is TPU. That is TensorProcessingUnit.
TPU is a dedicated chip designed specifically for machine learning applications. By reducing the computational precision of the chip and reducing the number of transistors required for each computational operation, the number of operations per second that the chip can run can be higher, so that a finely tuned machine learning model can run on the chip. Faster and faster for users to get smarter results. Google embedded the TPU accelerator chip in the circuit board and used the existing hard disk PCI-E interface to access the data center server.
6, Zhongxingwei
In China’s chip industry, which relies heavily on foreign imports, China Star Micro can be described as a “dark horse†that stands out. In June 2016, Vimicro first introduced China's first embedded neural network processor (NPU) chip "Star Smart One", which is also the world's first embedded video acquisition compression coding system-level chip with deep learning artificial intelligence. , and has achieved mass production on March 6.
The chip adopts a data-driven parallel computing architecture. The power consumption of a single NPU (28nm) is only 400mW, which greatly improves the ratio of computing power and power consumption. It can be widely used in smart driving assistance, drones, and robots. And other areas of embedded machine vision.
7, Microsoft
Microsoft stayed for six years and created a chip to meet the AI ​​generation. That is ProjectCatapult. This FPGA now supports Microsoft Bing. In the future, they will drive new search algorithms based on deep neural networks—artificial intelligence modeled on the human brain’s structure—in the performance of several commands of this artificial intelligence. Ordinary chips are orders of magnitude faster. With it, your computer screen will only blank 23 milliseconds instead of 4 seconds.
8, KnuEdge
KnuEdge is not actually a start-up company. It was founded by the former head of NASA and has been operating in an invisible mode for 10 years. KnuEdge recently stepped out of the stealth model and let the world know that they were getting a $100 million investment from an anonymous investor to develop a new "neuronal chip."
KUNPATH provides LambaFabric-based chip technology, and LambaFabric will perform neural network computing with completely different architectures than GPUs, CPUs, and FPGAs currently on the market. LambdaFabric is essentially designed to scale up to 512,000 devices in a demanding computing environment with a rack-to-rack latency of only 400 nanoseconds and a low-power 256-core processor.
9. Horizon Robot
Founded by Yu Kai in 2015, HorizonRobotics (Horizon Robots) has received undisclosed seed funds from investors including Sequoia and legendary venture capitalist Yuri Milner. Later, it obtained a joint investment that had already obtained Morningside, Takahata, Sequoia, Jinsha River, Linear Capital, Innovation Workshop, and the Real Fund. They are embarking on a one-stop solution for artificial intelligence that defines “everything intelligence†to make life easier, more fun, and safer.
Horizons is committed to creating an artificial intelligence "brain" platform based on deep neural networks - including software and chips that can achieve low-power, localized solutions to environmental awareness, human-computer interaction, decision-making and other issues.
10, krtkl
Krtkl was founded in 2015 to create "a tiny wireless computer to create something totally different." This development board is based on the Xilinx Zynq SoC, which integrates an ARM processor and a programmable FPGA. The user can even program it through a dedicated APP on the mobile phone for 230 user I/O interfaces. The application is smart and compatible with many expansion boards. Its features are as follows: Select Zynq 7010 SoCchip, integrated dual-core ARM Cortex-A9@667Mhz processor and 430KLUT FPGA resources (advanced to UT).
One of the highlights of this development board is that it not only supports traditional MicroUSB programming, terminal debugging equivalents, but also supports mobile terminal control, application of official applications, through the Wi-Fi connection development board, users can download programs, pins Holding, pin multiplexing and system management can be effective.
Shenzhen Kaixuanye Technology Co., Ltd. , https://www.iconline-kxy.com