For the most part, a positive outlook for the future of chips.
2019 will be the year of changes in the semiconductor industry, when new areas of research will spur technological breakthroughs
The 2019th began with cautious optimism about the semiconductor industry, despite all the black clouds that are gathering on the horizon. Such market segments as cryptocurrency and virtual reality do not meet expectations, the smartphone market looks saturated, DRAM prices are falling, which makes it necessary to cut costs. Companies involved in the automation of the design of electronic devices, talk about the suspension of sales in China in light of the trade war between him and the United States. There is a slowdown in the consumer electronics market, as evidenced by the latest earnings data from Apple.
But against the background of all this, some moments make us feel excited. Rapid AI Implementationfueling progress in areas such as the automotive industry and the Internet of Things. 5G, another important opportunity, is preparing for a major implementation. The number of new electronics under development is growing, which is generating ASIC recovery and the emergence of embedded FPGAs.
Most of the cryptocurrency fell by 75% during 2018, and no signs of stabilization have been observed so far. “Bitcoin continues to fall and may not recover in 2019,” warns Tom Vaughn, director of marketing and design for Cadence Design Systems . “This will seriously affect the production of silicon substrates, since a large number of them consume the cryptoindustry. Due to the fact that the cost of mining exceeds the value of Bitcoin, many miners have left the market. And this trend captures not only private traders. I predict that for industrial miners this will result in an even bigger problem due to their overhead (electricity, server infrastructure, buildings, etc.). I think some of them may close completely. ”
Another disappointment in 2018 was the augmented and virtual reality. “The problems with adopting AR and VR are that they are not yet suitable for serious immersion,” says Simon Forest, director of Connectivity & Connected Home for Imagination Technologies. “And the technology is not let down by graphics resolution or visual quality, but by the quality of working with it as a whole, which is still low.”
Thomas Urman, director of business development at the EV Group, agrees. “Content production is another problem,” he says. - The use of technology will be limited until the video transmission speed is high enough, and until we get to the acceptable values of the delay at which a person will not be swayed. It can replace a computer screen, but for widespread use, AR will have to wait until 5G is deployed, and it will take some time. ”
And this may not be all the problems. “The 360 ° video and spatial audio, which creates an effect of presence, looks promising in expanding the market beyond the needs of hardcore gamers, but we need sets of standards for video and audio, around which content providers could gather,” adds Forest. “In addition, we need standards that relate to human extra senses, touch, and smell. This will separate VR from a group of content formats with a weaker presence effect. Until then, VR will remain a niche topic, limited in use. ”
Leading Place for AI
There are several bright points in the semiconductor market that should accelerate growth and compensate for the fall in memory prices. “In particular, there is great hope for a critical increase in investment in new specialized processors, for example, for AI and machine learning,” said Wally Rines, honorary director of Siemens, Mentor. “For many applications in the field of pattern recognition and data analysis, chips with traditional architecture do not provide adequate speed and such energy savings as are required for the latest MO algorithms. As a result, the development of specialized chips has accelerated. This also stimulated the development of new chip design methodologies, many of which use AI to improve the process. ”
However, to believe that AI will disperse this industry, you need to be careful. "It is likely that the AI of many companies will enter the phase of 'getting rid of illusions' of the famous cycle of the hype of the company Gartner, - warns Forest. - The industry will quickly realize that the AI cannot answer all the questions, and the hype will take a look. Someone will quickly change the focus of development, ensuring the preservation of the useful capabilities of AI that complement existing systems, but AI will not necessarily remain the central pillar of these systems. ”
It is necessary to define expectations. “We all dream of AI devices or robots capable of executing orders in a free form, like“ wash the dishes ”,“ paint the fence ”or“ go to the parents' house ”, but I see that in 2019 the growth of the AI will be associated with simplifying routine tasks that can be automated by adding AIs, particularly in the areas of voice or image recognition, ”says Mark Greenberg, director of product marketing at IP Group at Cadence. “For example, if you don’t like cooking microwaves on a keyboard, voice recognition — as a separate application or using a virtual assistant portal — will allow you to order the microwave to warm up your food. We will see more office security systems using face recognition to open doors, and possibly
AI is still in its infancy, despite the speed of development. “There is a lot of talk about the dangers of AI, but they seem to be based on the concepts of singularity and ethics taken from science fiction,” said David Harold, vice president of marketing communications for Imagination. "Today, the use of AI is in the hands of people - and it is on these people, especially those who are empowered with legislative powers, we must pay attention, trying to ensure that the AI will help society, and not burden it."
For the implementation of AI, there are two very different areas: training and conducting logical inferences. Inference can occur either in the data center or on the client side.
AI in data centers begins to change significantly. "Today, Nvidia and Intel Xeon-based GPUs dominate the neural networks of data centers," said Geoff Tate, director of Flex Logix. - Tesla T4 from Nvidia was the only card in use for training data centers. Habana Goya will be the first product, whose cost and speed will significantly surpass Tesla T4. In 2019, new engines will be launched with an optimized process of logical constructions, and their introduction into data centers will accelerate, which will dramatically reduce the market share occupied by Nvidia. Own decision-making accelerators, for example, Inferentia from Amazon, will contribute to this trend. ”
And the trend is likely to continue. "The dominance of Intel in data centers will decline," adds Tate. - Specially designed for training and decision making engines will take a significant part of data processing from neural networks built on Xeon processors. Acquired by Intel neural networks is still to gain momentum. And this is critically important, since neural networks make up a rapidly growing share of computational loads in data centers, and [ordinary] processors cannot compete with chips optimized for neural networks. ”
Moving towards optimized architectures will allow implementing decision-making systems on the user's side. “Next year, we will see significant progress in developing adaptive, learning systems that can be used both in the cloud and in a little power-consuming, stand-alone applications on the user side,” says David White, senior director of the research and development group at Custom IC & Cadence PCB Group. “This will require a more effective combination of machine and depth learning with adaptation aimed at optimization.” It will also make it tighter to control systems that use AI in industry and transportation. This is not such a big problem as cloud marketing or image processing speed, but in the process of integrating AI and in-depth training in transportation,
Moving to the client side will take some time. “To implement this AI on the client side, we need to consider innovative methods for improving the packing density of transistors on silicon chips,” Forest notes. - We expect to see completely new ways of creating systems on a chip, which have both opportunities to gain new knowledge through training, and the necessary skills of logical constructions for adaptation. Several more years should pass before such opportunities appear. ”
The combination of AI and 5G will be a great help for the industry. “AI and 5G are the technologies needed to build robots,” said Lauro Rizati, a control consultant. - Also, these technologies will serve to create new gadgets from the field of the Internet of things, which will be used in many areas that are not very useful. AI and 5G will create even more business opportunities and the creation of new jobs for engineers than in 2018. ”
We can also expect network expansion. “In 2019, 5G will work as an upgrade for existing services,” says Forest. - The technology will inevitably start working first in the cities, and then for several years will expand the area of applicability when smartphones will include support for 5G wireless technologies. Companies will experiment with 5G to deliver data from the AI to the cloud, 5G modems are likely to appear among devices from the Internet of things, and there may also be a desire to create 5G modems with low power consumption to help this technology spread among products for the internet of things. "
In some cases, these technologies are being introduced too quickly. “In Tokyo, 5G was actively promoted, but they were ahead of the adoption of the standard,” Urman says. - We will see a lot of additions. It all happened too fast. This story should develop in 2020-2021, and then AR will enter the market. But for production on a large scale in three years, the technology should be ready today. To use AR, a physical connection is not needed - centers are needed, the opportunity to demonstrate premises is needed - content will play a key role here. ”
The technologies used in roboMobiles will change not only the movement of cars and trucks, but also the financial landscape associated with their purchase, maintenance and insurance, and as much as the technologies that will work under the hood or behind the dashboard. "Honestly, it is too early to talk about who will win the race and be able to offer an autonomous car first, but you can say that all major automakers and newcomers in this area are actively investing in development, and want to enter the market early," says Bryce Johnston , director of automotive segment of the company Imagination. “I put the fact that the first significant advances will occur in such commercial areas as robotaxi and truck convoys that attract the most public attention - when this technology works,
Someone wants to buy their own robo. "In 2019, we will see the switching of the ro-mobile industry to the upper gear when more pilot tests begin to take place all over the world," said Jan Pei, director and co-founder of Cepton Technologies. “Consumers will already be able to enjoy the active safety functions of the third level, as soon as it will be allowed by the officials - perhaps already in 2019. Among these functions - help in acceleration, taxiing and braking, which will further increase driving safety. With the increase in the number of orders for lidars from automakers and OEM-companies, companies engaged in lidar will experience a test of strength. We will see a serious consolidation of the lidar market, since many companies are not yet ready to produce them on such a scale as are required by the automotive industry.
With the advancement of the automotive industry, new demands are also being placed on chip design. “Manufacturers of auto parts require zero rejection of chips from suppliers to ensure that an increase in the share of electronic components in a car will not increase the number of defective cars,” says Art Schaldenbrand, the main product manager at Cadence. “Judging from the data on product returns, about 80-95% of these returns are due to analog elements. As a result, independent chip manufacturers supplying goods to automakers should acquire tools to analyze the coverage of their tests. In 2019, IEEE P2427, the standard for modeling and coating analog defects, comes into play, defining how model failures are involved in calculating test coverage. ”
In 2019, we will also see updates to ISO 26262. “The new generation of ISO 26262 came out in December 2018,” adds Schaldenbrand. “The ISO 26262-11 standard: 2018, standards for applying ISO26262 to semiconductors, provided the first specific standards for analyzing the safety of semiconductor circuits.”
The auto industry is making demands on an increase in the number of different security tools. “Functional security has become a major concern in the semiconductor industry and systems development, and their main applications are robomobils,” said Sergio Marcheze, technical marketing manager at OneSpin Solutions. “In 2019, we will see an increase in security issues requiring resolution in other areas. Homeowners and businesses do not want a missing bug or alpha particle to turn off alarms or sensors, making them vulnerable to invasion. ”
Internet of Things (IoT)
Over the past couple of years, the penetration [of this area into our life] has continued silently. “The coming year will focus on simplifying the development and large-scale implementation of vertically integrated solutions, specifically using IoT,” said Vivek Mohan, IoT director at Semtech. - An example is the retail sales industry, in this area we will continue to significantly invest in IoT technologies, both online and in physical stores, which will allow players in the market to win back more segments. These networks will be under pressure to improve efficiency in areas such as energy savings and standards compliance — food security, data privacy, worker security, and so on. ”
This will affect many aspects of the community. “In 2019, there will be a surge in the emergence of products for the smart home, intelligent control of IoT will manifest itself in areas such as home lighting, watering and heating / cooling, will increase the automation and efficiency of everyday tasks,” Arm predicts. - However, the development of machine learning and computer vision will allow the city authorities to engage not only in reducing costs (for example, installing smart LED street lighting), but also engaging citizens in the life of the city and raising funds from areas such as red traffic lights and WiFi access points , 5G services, smart skyscrapers, crime detection and analysis, broadcast information dissemination. ”
Power consumption is a big problem in many IoT applications. “In 2019, many will focus on environmentally sustainable power systems that will allow for the widespread adoption of IoT,” said David Soo, director of Atmosic Technologies. - To reduce the energy consumption of IoT devices, the industry will apply advanced technology development, scaling and improvement of technological processes, optimization and customization of design. As a result, we will increase the battery life. Improving the design will allow such functions as waking up on request to clarify sleep cycles and device operation, which will further reduce energy consumption. The ultimate goal of reducing consumption is a perpetual battery. In 2019, we will see devices with a battery in which they work during their entire lifetime. ”
Another important IoT issue is security. “Although in 2018 we saw the beginning of joint work to improve the security of semiconductor devices connected to the network, in 2019 even more companies will work together and provide simpler and more reliable solutions for the introduction of products and services,” says Martin Scott, technical director, senior vice president and general manager of cryptography at Rambus. “One of the ways to work together is to use open source and hardware. The open nature of RISC-V ISA can inspire innovation and promote widespread adoption, leading to the sharing of best practices. But we cannot limit ourselves to iron — there is a real need to improve the security of software and network connections. ”
To ensure the security of the SoC, a lot of work needs to be done. “During the development process, there should be no leakage of confidential information, and it should be protected from hacking and other external attacks,” says Marchese. - Trojans in the gland, such as switches or black moves, introduced with malicious purposes, become one of the main problems. Most SoC integrators and information providers do not take this risk seriously enough. We expect that in 2019, many companies will begin to get acquainted with this topic, and someone will learn new tools and processes that allow them to catch Trojans. ”