(Figure from Phoenix Technologies)
Recently, Japan, SOFTBANK decided to spend £ 23.4 billion acquisition of United Kingdom ARM chip design company company, the purchase price is equivalent to ARM the revenue 23 times last year, equivalent to 70 times net profit in 2015. Purchase motives, media claim that "Japan SOFTBANK tried to acquire ARM associated with artificial intelligence", which he swears "as to why the acquisition ARM, they will have to start from SOFTBANK Chief Executive Masayoshi Son, the dream of artificial intelligence. In recent years, SOFTBANK has been actively promoting the development of robotics, they offered Pepper humanoid robot, and will continue to enhance the company's artificial intelligence industry competitiveness. "
So the truth is, as the media say it?
| ARM's main business are there?
ARM company is the world's IC design giants, ARM processor occupies about 95% tablets and smart phones. Direct sells chips and Intel, ARM does not directly sell chips but sell technology licensing. ARM product line is very long, the system IP, physical IP, GPU, video decoding and display products. ARM Cortex family of processors is the most widely known. Which Cortex-A for open systems; Cortex-R for embedded systems Cortex-M for various micro-controller SC series, electronic government, SIM cards and other security-oriented market. These products are the cornerstone of ARM and Intel rival.
In recent years, the ARM is actively expanding low-power server market, although it has not yet opened in the server market--on the software ecosystem still failed to form a climate, also d-series Intel Xeon chip in hardware attacks. But in the area of artificial intelligence, ARM is a lot of action. According to the company's earnings report revealed that there is a Blue Sky Program within the company, only in the 2015 fiscal year, 488 engineers into the ARM deployed "ARM products had never been involved in" research and development, and have not been involved in this product area is artificial intelligence. Japan SOFTBANK's conditions will dramatically increase the number of employees of ARM, which expanded employee will also invest in the research and development of artificial intelligence. Therefore, if the ARM has always been regarded as confined to the embedded chip IC manufacturers, would be OUT.
| Acquisition of ARM robot industry and development Juicy Couture sale
Robot has three big core parts, respectively is reducer, and servo motor, and controller, three who respectively accounted for robot machine cost of 35%, and 25%, and 15%, and reducer and servo motor market share most was developed by occupy--in reducer aspects, Japan Harmonic is harmonic reducer field of leading who, occupy has about global 15% of market share; Japan Nabtesco in RV reducer field of global market share about for 60%. Servo motor, Japanese companies around the global market share of 40%; Siemens, Bosch, Schneider and other German brands global market share of about 30%.
(The Godson robot controller)
Controller equivalent to the robot's brain, for publishing and delivery instructions, including two parts: hardware and software:
Part, the control algorithm of the software. Hardware is industrial control board, including some of the main control unit, signal processing circuit. Real ARM and hitch is the controller of the chip, and ARM some market share at the controller, but the controller is not the only ARM. In fact, robot controllers based on Intel X86 chip can do, even a Loongson-based robot controller, Furthermore, there are industrial robotic controller based on DSP and FPGA.
Therefore, the robot controller, X86, ARM, MIPS is in the same stage of the era, there is no X86 in the area of PC, ARM in the case of mobile phones, flat areas are in a monopoly position. Therefore, said buying ARM for the development of robots, is somewhat far-fetched. But the said ARM robot industry and development business, it would be biased. After all, in terms of robot controllers, ARM still holds a considerable market share.
| What chips are available for artificial intelligence?
AI is calculated as a means to achieve human-like perception or cognition, problem solving or research derived. Words simple machine is used to simulate people's way of thinking, so that computers can be carried out in a manner similar to the human brain thinking, speech recognition, identification, understanding the meaning of the image.
Currently two core factors of restricting the development of smart:
First, the algorithm has not reached perfection, now the insufficient understanding of the process of intelligent computing.
Second, hardware performance is not enough. There are billions of neurons of the human brain, complex network consisting of baiwanyi synapses, and existing chip and that there is more than one order of magnitude difference.
In order to develop a sufficiently powerful hardware performance, researchers used three types of practice: research and development of neural network chip, DSP, and the GPU.
Developing hardware for artificial intelligence, the first approach is to develop special chips used for artificial intelligence, neural network chip is. Neural network chips can be used to solve machine vision, hearing, and practical problems in the areas of natural language processing, can be used for Convolutional neural networks (CNN), the recurrent neural network (RNN) and short-term memory (LSTM) of speech and natural language processing networks, the most typical representatives of China's "Cambrian" chip, IBM's "true North".
(Cambrian)
Second species practices is will traditional of oriented digital signal processing of DSP processor schema for processing neural network (main in operation device aspects for has corresponding modified, for example low width and beyond function), this with traditional SIMD/DSP schema to suitable distribution neural network of technology thought in international Shang has has many precedent, even has mature of products, for example China of "stars intelligent first,", and CEVA company of XM4 processor, and Cadence company of Tensilica Vision EV processor P5, Synopsys Corporation. This DSP supports only neural network into operation to support training of neural networks, compared with neural network chip has some gaps.
Third species practices is used GPU, most typical of example is inve up for artificial intelligence custom of GPU, but this GPU and neural network chip compared, has times above of energy disadvantage--to "Cambrian" of DianNao for cases, DianNao for single nuclear processor, frequency for 0.98GHz, peak performance up each seconds 452 billion times neural network basic operation, 65nm process Xia power for 0.485W, area 3.02mm^2. In a number of experiments on the neural network representation shows that the DianNao average performance and mainstream GPGPU, but size and power consumption of only mainstream GPGPU 1% magnitude.
Then the ARM of your choice of practice? According to sources in, ARM is already working on NPU and DSP, it is said that after the SOFTBANK acquisition ARM, new employees will also invest in research in this area. Future, the ARM chip may be integrated in the phone just CPU, GPU, DSP will also integrate NPU and artificial intelligence-related services provided to the public.
| Conclusion
Google "Alfa Laval dog" after World War Li Shishi, AI was hot, seems to have become the darling of the capital markets, which is China's "Starlight intelligence first" belongs to DSP, but claimed to be an important cause of neural network chip. In the case of recent negative news by, released to the artificial intelligence to enter the amounts on the capital market a positive signal--after all, "no money is absolutely indispensable" by some media and capital operation, upgrade their share prices are also helpful for SOFTBANK. Juicy Couture iPhone 6 Plus Case
Lei feng's network (search for "Lei feng's network", public interest) Note: author tieliu, micro-tieliu1988. Lei feng's Web exclusive article, reproduced please contact the authorized and reserve the author and the source, no deletion of content.
No comments:
Post a Comment