Form： laoyaoba.com 2018/3/8 Browse：2650 Keywords: Mobile phones chips Qualcomm Smart Huawei Apple Solutions companies Technologies products Development Design Market Integrated Circuits Microelectronics Semiconductor ICs Mobile phones
Handset chip maker Qualcomm has now launched three generations of platforms aimed at artificial intelligence (AI) and continues to focus on AI applications and the hardware and software needs that they require. As AI continues to grow in the handset industry, its competitors are gradually increasing. Mobile phone brands such as HUAWE and Apple have also been involved in the development of this sector, and related solutions (such as Kirin 970 A11 Bionic). In this regard, Qualcomm said that compared with Huawei, Apple and other mining hardware AI development strategy, the company still take the "hardware and software" approach, in response to AI development.
Gary Brotman, director of product management at Qualcomm Technologies, a Qualcomm subsidiary and Qualcomm Technologies, Inc., said Qualcomm has released three generations of platforms aimed at AI and has been concerned with AI application cases and the hardware and software requirements they drive. For the time being, the company does not think it is necessary to use proprietary AI hardware in its current use case, of course, that does not mean Qualcomm will not consider the development of dedicated AI hardware in the future.
Brotman further explained that the current challenge for the industry as a whole is that AI develops very rapidly and algorithms evolve almost daily. Therefore, the use of dedicated hardware platform has a certain risk, because in the development of the early need to predict the development of the field of AI in the next 18 months trends and hot spots, and the corresponding technology into a dedicated hardware platform.
In other words, the general hardware from the design and production to the market, the cycle takes about 18 months, and the AI algorithm evolves very quickly, when the hardware really on the market, the previous solution may no longer fit. In addition, the development of dedicated AI hardware will also increase the corresponding chip costs.
Brotman shows that most developers now use custom run-time layers in their AI algorithms, but the specialized hardware is not efficient enough to handle these custom layers, so developers have to call other modules such as GPUs or DSPs deal with.
However, on the software side, software tools from either ARM or other vendors offer very high levels of flexibility. Developers not only have access to the underlying hardware, but they also show that they have the flexibility to take full advantage of hardware capabilities through programming in some situations, such as the Snapdragon platform's DSP, to reduce the risk of developers altering algorithms. Therefore, Qualcomm in the future development of AI, will still be a combination of software and hardware to deal with artificial intelligence needs.
For the future layout, Brotman revealed that in the coming months, the company will release a benchmark system for AI performance, focusing on the following three aspects. The first is the performance of the AI processing case, including speed and number of frames; followed by power consumption; and finally precision. Qualcomm hopes to strike a balance between those three points, and if blindly pursuing the efficiency of processing, its accuracy may decline, so I hope to have a more balanced evaluation system.
Brotman said that, in fact, compared with the benchmark test, the application of the case is the real touchstone, AI performance can be the most real situation.