The semiconductor industry has a good development in 2017, and the market sentiment is still satisfactory to many manufacturers. 2018 The advent of the new year 2018, the development of the semiconductor technology industry will be? What is the market mainstream of the key technologies? How the layout of the industry? The next year's market? The following will come together.
If you look at the overall industry, Gartner listed the top ten technology trends in 2018, including artificial intelligence (AI) -based, intelligent applications and analytics, smart objects, digital avatar, cloud to edge computing, interactive platform , Immersive experience, blockchain, event-driven, and continual adaptation to risk and trust.
Gong Peiyuan, Gartner's senior partner in Greater China, said the top ten technology trends may be somewhat similar to those in 2017, but it also means that each technology is getting more and more mature and further influencing industry trends and people Of life, but also become the focus of 2018 technology.
AI influence not only deep but also full of life around
Gartner proposed ten major technology trends, in fact, and Intelligent Digital Mesh (Intelligent Digital Mesh) is closely related to Gong Peiyuan explained that part of the smart is undoubtedly artificial intelligence, in 2017 the industry as the first year of artificial intelligence development, to In 2018, Gartner believes that artificial intelligence will move into the field of practical technologies into useful technological achievements rather than the technologies studied in the laboratory, and 2018 will be a year of artificial intelligence.
As a result, artificial intelligence will make a big splash in chip offerings, such as AI on Chip, by 2018. For example, the technology industry has begun to have a greater interest in servers based on Graphics Processors (GUPs); Google Server center to create its own proprietary chip Tensor Processing Units (TPU); Microsoft (Microsoft) released FPGA-based low latency Deep Learning Cloud Platform Project Brainwave; Intel developed specifically for machine learning Nervana chip; Nvidia's graphics processor GPU, and artificial intelligence chips; Qualcomm has introduced a special neural network chip; TrueNorth, the first brain chip developed by IBM; and Spiking, a neuron adaptive processor developed for unsupervised learning of brain chips, and Cerebras startup Groq expects to launch artificial intelligence chips in 2018.
Gong Peiyuan believes that AI has begun to 'proof' the future of the general public or the technology industry to have far-reaching implications in order to attract IT or chip companies to actively introduce artificial intelligence-related chip products to start-ups. For example, Artificial intelligence will create new applications in 'user experience' in 2018, meaning new user experiences that have been brought out by the Apple iPhone X in 2017 and will subsequently augment user data with artificial intelligence Analyze the process, reduce blind spots, and increase the ability to analyze specific aspects, giving users a new and different experience, and AI can also help service providers to provide better application services to meet the needs of consumers.
In addition, mainstream applications of artificial intelligence will emerge in 2018, especially in relation to 'security.' As artificial intelligence needs to gather more and more data as a basis for its in-depth analysis, artificial intelligence not only needs to protect data Security; more must be able to prevent and find out the wrong or false information, so as to avoid the analysis of the error.
Not only that, another important point in the development of artificial intelligence in 2018 is the interaction between people and artificial intelligence.Gong Peiyuan shows that people still have doubts about artificial intelligence at this stage, in particular, artificial intelligence will not replace artificial, in fact, , If people or businesses truly understand 'using' artificial intelligence, not only will not be replaced, but also receive 1 + 1 greater than 3 synergy.
This means that, from now until 2019, artificial intelligence will replace more jobs than it creates; however, by 2020, Gartner predicts that the opportunities created by artificial intelligence will be enough to make up for its replacement, not to mention that artificial intelligence The productivity of many jobs will be raised. If skillfully applied, it will indeed enrich people's employment life, remodel old jobs and create new industries.
Institute of Information Industry Information Technology (MIC) for the 2018 ICT ten trends in the industry, there are several and artificial intelligence related to the policy of the MIC Industry Consultant and Director Hong Chunhui said the development of the top ten in 2018 trends In fact, around the four directions of intelligence, opening up, service and integration, among them, the development toward the intellectualization of industries has a profound relationship with the development of artificial intelligence.
In 2018, artificial intelligence will be able to get out of the dilemma of less data to be analyzed in 2017 and enhance the capability of algorithms. This will lead to breakthroughs in technologies such as voice and video identification, and will further increase the added value of intelligent terminals and develop more intelligent Applications Retail services can also benefit from the evolution of AI, analyzing customer attributes and developing services that better meet customer needs, as well as the demand for intelligent terminal devices with video and audio capabilities.
In addition, artificial intelligence technology can also make the capital security system more complete.Hong Chun-hui believes that from outside the system intelligence analysis, endpoint behavior anomaly analysis, authentication uses artificial intelligence to strengthen the ability to create more efficient, more instant As the number of Internet of Things (IoT) devices increases rapidly, the development of artificial intelligence applications for systems and products will also become the key driver of AI's AI.
How to grasp the business opportunities?
According to Gartner's estimates, deep neural network (DNN) and machine learning applications will create $ 10 billion in market opportunities for semiconductor suppliers by 2020, and artificial intelligence and machine learning will gradually penetrate everything to become a technology vendor for the next five years The main battlefield.
And how to grasp the artificial intelligence business in 2018 or even the next few years derived from the huge business opportunities? Gong Peiyuan said that the digital avatar, intelligent Internet of things agents, blockchain, mature artificial intelligence and other technology organizations from intelligent digital grid derived The new business solutions will be more realistic than ever before, and neither the high-tech industry nor end-users should ignore them, and more practical applications and solutions will emerge in the coming year.
Gartner believes I believe companies have witnessed and felt the maturity of technologies such as artificial intelligence, the Internet of Things and blockchain in 2017, and the obvious innovation in artificial intelligence-based business models and their solutions. As a result, global CIOs Business will return to breakthrough innovations led by smart tools and the App All high-tech industries should begin to face, plan and adopt AI technologies at this stage, clearly defining the 'smart' scenarios and the business-level implications At least until 2020, the main battlefield for technology makers will be how to build systems that learn, adapt and may even act autonomously.
Semiconductor / electronics industry should be locked in the direction
Artificial intelligence in 2018 years or even the upsurge in the next few seemingly will not be interrupted, in addition to the general business in addition to import artificial intelligence from their own internal, but also should start to use artificial intelligence to create closer to consumers. Gartner predicts that by 2022, will More than 80% of IoT projects launched by enterprises will have built-in artificial intelligence components, representing a substantial growth compared to less than 10% in 2017. How can semiconductor manufacturers grasp the opportunities in artificial intelligence?
Gartner Research Director Zheng Yajun said that Gartner's top ten technology trends have the semiconductor industry can cut into the place, which in response to the popularity of artificial intelligence, the future computing architecture will be a great change, and the artificial intelligence and the Internet of Things is bound to further And more advanced computing architectures will also drive the emergence of heterogeneous computing architectures, the persistent emphasis on persistent memory, the increased demand for solid state drives (SSDs), and the adjustment of flash memory arrays (AFAs) For solid state arrays (SSAs), high-performance networks are imperative, and higher-density packaging is even more important, so the relevant semiconductor manufacturers can cut into the above points, the introduction of the corresponding products or new technologies.
Zhou Shih-hsiung, consultant and deputy director of the MIC industry at the Policy Committee, pointed out that the development of artificial intelligence will accelerate the development of intelligent objects. The future semiconductor industry will not only be able to research and develop artificial intelligence chips, but also will have artificial intelligence inference and training. Therefore, A wide range of ASICs, such as voice-recognition ASICs, image processing ASICs, motion control ASICs ... will flourish to assist and share CPU or GPU workloads and be tailored to the needs of the application's endpoints The product.
More importantly, when designing ICs, digital twin thinking can not only reduce the cost of computing architecture due to artificial intelligence, but also allow Taiwan's semiconductor industry to develop ICs in the future Products, you need to think more about the part.
On the other hand, the combination of artificial intelligence and the Internet of Things will also lead to the emergence of many intelligent IoT objects, which are physical objects that can perform tasks according to a fixed programming model and make use of artificial intelligence to make more advanced behaviors, meanwhile Can more naturally interact with humans in its immediate surroundings and with humans.Artificial intelligence is driving the technological advances of new smart objects such as autopilots, robots and unmanned aerial vehicles and also enhances many existing objects such as Internet of Things Consumer and Industrial Systems.
Cheng Yajun reminded that future smart objects such as smart phones, which are indispensable to people, will certainly have a built-in Natural User Interface (NUI) that is smarter than the user, so the internal chip architecture of the mobile phone will also be different. Rely on a main CPU for natural user interface and artificial intelligence, or other parts of the operation, will be stretched, it is necessary to build another co-processor (co-processor) In addition, the integration of the natural user interface Smart objects must also incorporate more sensors, and the semiconductor industry needs to ponder how to further reduce the power consumption of the main processor and the overall power consumption of the device.
Zhou Shixiong said that natural user interface including natural language, face, iris ... identification, the future of these human physiological characteristics of identification technology, will be more accurate due to the progress of artificial intelligence.However, to achieve the natural user interface on the smart objects, In addition to sensors, microphone arrays, audio or video processors, video cameras ... and so on, all offer new business opportunities for semiconductor manufacturers.
Strong memory market roll
In addition to the main processor or sensor, dedicated ASIC, the development of memory is also quite noteworthy changes in the cloud services, data centers, computing architecture, intelligent objects derived from the popularity of artificial intelligence.Zhou Shixiong that the memory The key to subsequent product development will be high-bandwidth data storage and transmission, with high transmission efficiency and low latency due to next-generation memories such as magnetoresistive random access memories (MRAMs) and variable-resistance memories (RRAMs) To meet the needs of smart applications with high capacity and high bandwidth. Even though the price is slightly higher, some artificial intelligence or Internet of Things applications are less cost-sensitive, so they took the lead in adopting next-generation memory and boosting the demand of next-generation memory.
According to the statistics by the MIC, the next-generation memory currently accounts for only 0.15% of the total memory shipments, and is expected to grow to 3% by 2020 and to reach 139.8% CAGR for next-generation memory from 2015 to 2021. Zhou Shixiong Pointed out that in the future, the application market will continue to improve the miniaturization and performance requirements of the memory process. For example, the high-speed embedded memory is required for the car and the low-power embedded memory for the IoT product. Thus, the development of embedded memory will be 2018 for the memory supplier Year or future, and foundry and package makers such as Samsung and TSMC have also started to develop eMRAM and eRRAM technologies for embedded memory needs.
On the memory market trend side, DRAMeXchange research manager Huang Yu-hsuan said that in terms of applications, smartphones have always accounted for the largest category of memory shipments due to the introduction of artificial intelligence in mobile phones to be smarter than humans or to display 4K or 8K image quality, or the implementation of other emerging applications, the application processor performance has to continue to increase, directly contributing to the memory capacity also increased, which also allows the average DRAM DRAM growth in 2017 has more than 40% 2018 DRAM The price will still go up, showing the "no highest, only higher" situation.
Memory requirements can not be ignored is a black horse server.Huang Yu Xuan believes that the built-in artificial intelligence in mobile phones and the cloud to be combined, this time is the need for server computing instant support, the entire framework complement each other to provide users with the best The service or function of the current smart phone built-in mainstream memory capacity of 8G, but in 2018 server memory capacity is expected to see 10G, and no capacity ceiling is expected in 5 years, smart phones will be the highest adoption of DRAM s installation.
DRAMeXchange analyst Liu Jiahao said the main driving force for server storage demand rising from the cloud, data centers, Internet data streaming and enterprise server room for server performance requirements increase.Among them, the construction of data center server demand This is because the amount of data to be transmitted and processed in the future will continue to skyrocket and the data center will have a higher demand for servers with a single workload and high performance computing in order to handle large amounts of data, To achieve more applications.
Server market can not be ignored
Smartphones and personal computers (PCs) will perform roughly the same as in 2017 by 2018, but servers can be a target market for developers looking to find alternative ways. "Liu Jia-hao believes that in addition to the above-mentioned demand side, Arm camp Actively enter the server market, you can see the server for the semiconductor industry, the future will be an important business opportunities, although Taiwan's foundry industry has stuck to a good place, but still be cautious to deal with market changes.
Part of the server's key processor components, Liu pointed out that X86 architecture is still the mainstream market in 2017, Intel (Intel) and AMD in the server processor market share is close to 100% 2018 is still the case, only Intel and Intel AMD respective market share has declined.
In addition, the architecture of the server main processor platform will change slightly as the GPU has begun to enter the server main processor market to provide accelerated computing. The server chip built by Google includes an ASIC, but because of the inability of the ASIC to adjust, Its own server platform will be equipped with FPGA, the future server main processor built-in chip FPGA will surpass ASIC.
Industry consolidation will continue
Finally, Hong Chunghui said that no matter how the semiconductor or ICT industry is affected by artificial intelligence issues in 2018 and moves towards intelligence, the continuous industrial consolidation wave in recent years will continue to occur again in 2018. For example, the most watched by the end of 2017 Of Broadcom want to buy Qualcomm news, although at this stage 'conditions' have not talked together, will eventually happen in 2018, remains to be seen.
In addition, in addition to the continuous wave of integration among chipmakers, service providers and telecom operators are also moving in the direction of industrial concentration and the follow-up effect of mergers and acquisitions is also a matter of concern for Taiwan's ICT or semiconductor industry. According to Hong Chunhui, Broadcom, if successfully acquired Qualcomm, will occupy the majority of Netcom chip market, but the Taiwan IC design industry may still benefit from the customer transfer efficiency, and have good opportunities for development.