Global semiconductor companies to keep up with the development of in-depth learning innovation
Author: huifan Time: 2017-08-18
The slowdown in market demand and improved performance are driving renewed wave of innovation in the global semiconductor industry. Companies in the industry are struggling to develop new chip designs, materials and manufacturing processes, one of the reasons is that deep learning of this artificial intelligence technology is More and more widely used in picture classification, voice translation and automatic driving and other tasks.
Computer chip makers are struggling attendance management to cope with some of the toughest times in the history of the industry, as demand for certain devices slows and the role of smaller circuit systems in performance improvement continues to wane.
However, industry executives say these pressures are driving innovation in the semiconductor industry and creating new startups interested in harnessing the current dilemma.
In the $ 350 billion global semiconductor industry, large and small companies are working to develop new chip designs, materials and manufacturing processes. One reason is that artificial intelligence, called depth learning, is increasingly being used for tasks such as picture classification, speech translation and autopilot, which benefit from new computer technology.
Some of these new development efforts directly to subvert Intel (Intel Co., INTC) and other well-established veteran as the goal, Intel has been through the adjustment of some of its proven strategy to respond.
Dharmendra Modha, chief scientist at International Business Machines (IBM), says it is the best of times and the worst. He is currently leading an unusual project to develop chips simulating the human brain.
At present, semiconductor companies are particularly active in the field of deep learning. The technique involves training the system by exposing the system to massive amounts of data, unlike programming a system with explicit instructions, which are not only time-consuming but often less reliable. Internet companies using depth-of-learning technology are particularly interested in promoting innovation in hardware that can lead to faster results.
In-depth learning systems typically use both Intel processors and chips from Nvidia Corp. (NVDA) or Advanced Micro Devices Inc. (AMD), which were originally designed to present video game screens. These chips contain hundreds of thousands of simple processors that can run simultaneously, while the Intel high-end chips contain dozens of complex computational cores.
Some companies even believe that it is necessary to develop more specialized hardware. Alphabet Inc. (GOOG) (Google) recently took an unusual step, from scratch to design chips to meet the needs of some of the depth of learning tasks. IBM introduced TrueNorth in 2014, which was developed specifically for in-depth learning. The chip contains one million simulated brain distributions. Modha said the chip has shown surprising rapid progress in depth learning applications Fingerprint car lock and is expected to create a large-scale business by 2019 as planned.
Venture capitalists also take note of this. The chip industry's technical challenges and fierce competition lead key car lock to most venture capitalists hid access control system to spend money elsewhere. But some entrepreneurs and investors see new opportunities to develop chips for markets such as networking, where consumers may want to diversify their sources front door lock rather than rely on one or two major suppliers.
Start-up company Cerebras Systems founder Andrew Feldman said the company found it was easy to raise funds and raised funds in eight days. But he did not disclose the amount of funding. The company has 25 employees and plans to design a processor that is targeted for deep learning.
Other start-up companies to design deep learning chips include KnuEdge Inc, Graphcore Ltd, Cornami, and Wave Computing. Wave said the company is developing a set of professional processor system can be completed in 6.75 seconds to complete a typical text analysis work, while equipped with Intel Xeon processor and NVIDIA processor system to complete the same work is time-consuming 69 minutes.
At the end of 2015, Intel bought Altera Corp. (ALTR) for $ 16.7 billion to sell FPGA chips to companies such as Microsoft Corp. (MSFT). Intel in 2016 acquired the artificial intelligence startup Nervana Systems, and plans to the latter's in-depth learning technology into their own processors.