-
- December 30, 2022
- PCB news
- (0)
- 12 mins
10 technology trends of the global semiconductor industry in 2023
At the beginning of 2022, when the entire electronics industry is still in a shortage of chip supply, no one could have imagined that the reversal would come so quickly. In the second half of the year, consumer electronics markets such as PCs and smartphones have successively handed over data of severe declines in shipments. The weak downstream demand has also directly affected upstream chip companies and wafer foundries. The entire semiconductor industry has “wintered overnight”.
After entering the callback period, the growth rate of global semiconductor sales and demand has slowed down significantly. Wafer manufacturers in the industry that are chasing growth and expansion have significantly reduced their capital expenditures, and the semiconductor industry index has also retreated sharply in the second half of 2022.
At the same time, the United States’ anti-globalization technology blockade has further stimulated the determination of other countries to break through in advanced semiconductor technology. R&D in emerging technology areas such as chiplets, advanced packaging, high-performance computing chips, silicon photonics chips, and wide-bandgap semiconductors has gained success frequently.
AspenCore’s global analyst team communicated with industry experts and manufacturers during the year, and after summarizing and analyzing, selected 10 major technological trends that will appear or develop rapidly in the global semiconductor industry in 2023.
The metaverse ecology grows rapidly
Industry giants represented by Apple, Meta, Sony, and PICO will launch more AR/VR products in 2023 after early technology investment and precipitation. In particular, the addition of Apple MR devices will further accelerate hardware iterative upgrades.
In terms of hardware, silicon-based OLED technology is more mature, and the overall production capacity will be further improved. It will perfectly cooperate with the Pancake optical solution that is gradually becoming mainstream, bringing better comfort and immersion; from the perspective of cost, Pancake+Mini LED backlight Fast-LCD combination will still be the mainstream hardware solution.
In terms of data application, based on rich content and scene data, Metaverse will be deeply integrated with artificial intelligence, digital twins, blockchain and other technologies to reconstruct real-world products, processes or operations through digitization and virtualization, replacing or assisting people to play some constructive roles, such as safety training, industrial design and manufacturing, etc.
Chiplet has reached the inflection point
Moore’s Law is gradually approaching the physical limit, the production capacity of advanced chip manufacturing is in short supply, and the global environment is tense. Various factors have made Chiplet technology with high performance, low power consumption, high area utilization and low cost advantages gain more and more attention. According to Omdia data, it is estimated that by 2024, the global chiplet market will reach 5.8 billion US dollars, and by 2035, this figure will grow to 57 billion US dollars.
The popularization of Chiplet technology still faces many technical challenges, including not only physical electrical process/configuration, interposer and wire materials, communication interconnection/I/O, analog/logic design rules, power supply and signal control challenges, but also ecological and manufacturing challenges.
Considering that the ultimate goal of Chiplet technology is to be integrated into a “system-on-a-chip”, this requires high-density interconnection between each chiplet in order to achieve a high-speed interconnection similar to the functions of the original single large chip. Signal transmission speed between modules, is not an easy task.
The application value of silicon carbide is highlighted
Due to the high efficiency and high power density of silicon carbide devices, strong demand in new energy vehicles, solar photovoltaics, 5G communications, smart grids and other fields is expected to drive a rapid increase in the penetration rate of silicon carbide in 2023. In the field of new energy vehicles, SiC power devices such as on-board inverters, on-board chargers, DC/DC converters, and charging devices will enter a rapid outbreak stage with the electrification of vehicles and the large-scale use of 800V high-voltage platforms.
The demonstration effect brought by leading new energy car manufacturers such as Porsche, Tesla and BYD will drive more car companies to gradually adopt SiC solutions, especially deploy 800V high-voltage fast charging solutions on more high-end models to improve battery life Mileage, accelerate the arrival of the era of 800V architecture.
In the field of solar photovoltaics, benefiting from the “dual carbon” rigid policy goals and superimposed energy transformation needs, silicon carbide devices will be used in a wider range of photovoltaic inverters to prolong the life of the inverter and reduce the cost of replacing the photovoltaic inverter.
In addition, based on the consideration of reducing application costs, the research and development of large-size silicon carbide will develop from 6 inches to 8 inches. It is expected that the mass production time of 8-inch silicon carbide substrates will appear in 2023, and the industry is expected to usher in an explosive inflection point.
The high-performance computing completely turned to accelerated computing
On the TOP500 list for the past two years, players using NVIDIA GPUs and DPUs are frequently seen. At this year’s GTC Spring, Nvidia summarized the five major application areas of supercomputing: edge, HPC + AI, simulation, digital twins, and quantum computing.
It is almost inevitable that accelerated computing hardware will become the protagonist in the HPC field, because all of these applications have an insatiable demand for large-scale parallel computing. Accelerated computing hardware carriers include GPUs, DPUs, and AI chips at the top, and FPGA and more specialized chips at the bottom.
The rise of dedicated chips in the form of “DPU” is due to the main value of HPC and supercomputing, because GPU and AI chips are the main computing power, and DPU needs to provide a new carrier for the performance “aggregation” and maximized performance of different nodes.
The existence of DPU itself shows a new trend of accelerated computing. Coupled with the trend of the “low-carbonization” era, chips in the form of CPUs are gradually being marginalized in the high computing power market. The existence of CPUs like Grace Superchip is more for compatibility with old FORTRAN applications.
The golden age of 3D heterogeneous packaging is coming
3D heterogeneous packaging refers to the integration of separately manufactured components into a higher-level system-in-a-package in the form of Lego building blocks to improve overall performance.
In this definition, component can be any unit, such as a single chip, MEMS device, passive component, or even a subsystem, involving many key technologies such as materials, component types, circuit types, process nodes, and interconnection methods. It needs the support from multiple links in the supply chain such as foundries, packaging factories, EDA manufacturers, and material manufacturers.
However, the system-in-package is not just to package several chips together, but must meet the following conditions: 1. The volume must be reduced after packaging; 2. Different types of packaging technologies must be integrated; 3. It must contain various types of active and passive components.
According to the latest data from Yole Developpement, the total revenue of the advanced packaging market represented by 2.5D/3D heterogeneous packaging will reach 32.1 billion US dollars in 2021, It is expected to reach USD 57.2 billion in 2027 at a compound annual growth rate (CAGR) of 10%. The megatrends of 5G, automotive infotainment/advanced driver assistance systems (ADAS), artificial intelligence (AI), data centers, and wearable applications continue to drive advanced packaging forward.
Among the various sub-categories of packaging, the 2.5D/3D packaging market has the largest compound annual growth rate, which will increase from US$6.7 billion in 2021 to US$14.7 billion in 2027, up to 14.34%.
Overall, the packaging/assembly business is undergoing a paradigm shift, traditionally the domain of OSATs and IDMs, and is facing challenges and encroachment from different business models, foundries, and substrate/PCB Assembly supplier.
Integrated storage and computing solutions are coming out one after another
In 2022, the integration of storage and computing will usher in a small explosion in both theoretical research and substantive products. At the 2022 ISSCC, SK Hynix, TSMC and other major international manufacturers published papers on in-memory computing memory IP. SK Hynix also published the results of DRAM in-memory computing based on the GDDR interface, and demonstrated its first A sample of computing technology product GDDR6-AiM.
In October, SK Hynix also launched the industry’s first CMS that combines computing functions with CXL memory; Samsung disclosed its HBM-PIM solution at 2022 Hotchip, which is the industry’s first Programmable PIM solutions, Myhtic’s integrated storage and computing analog AI chips have also made new progress.
From the current point of view, most manufacturers develop based on SRAM, which is easier to obtain and lower in tape-out cost, and some are based on DDR. However, these volatile memories cannot save data after power off, and still face the problem of data transmission.
We predict that in the future, based on new types of memory such as MRAM and ReRAM, it will be the ultimate form of integration of storage and computing. They have orders of magnitude advantages in major indicators such as computing power potential, computing power accuracy, and computing power efficiency. However, the technology of these new memories is still immature, and we can only expect breakthroughs in the next few years.
ADAS advanced Level 3, primary automatic driving is on the way
The intelligentization of new energy vehicles has become a mainstream trend, and its core is to realize automatic driving. There are 5 levels of autonomous driving. At this stage, Level 2 ADAS (Advanced Driver Assistance System) is the main one. By 2023, the ADAS market size of Level 3 will gradually increase, and by 2027, it is expected to be equal to the market size of Level 2, and will continue to erode the sales of Level 2 cars, becoming the most common level of autonomous driving vehicles in the world.
Higher levels of autonomy mean cars must make fast enough decisions, and their R&D spending and barriers to entry are higher. In addition to core AI chips, autonomous driving solutions should also include connectivity, sensing systems, image training models, ADAS map development, route planning, vehicle control, driver monitoring systems (DMS), natural language processing (NLP) and smart cockpit solutions plan.
Self-driving chips must also be able to provide custom and region-specific algorithms. These need to be realized through cooperation between automotive OEMs and autonomous driving chip companies.
However, at present, with the decline in the cost of ADAS sensors such as cameras and radars, Counterpoint predicts that the penetration rate of global ADAS in car shipments will reach 78.7% by 2024. When electric vehicles take off in an all-round way, the advancement of ADAS will accelerate and fully penetrate the market.
Mobile satellite communication is rapidly popularized
In the autumn of 2022, Huawei released the Mate50 with a satellite communication function before the Apple New Product Conference, which set off competition in the mobile phone satellite communication market. With the basic popularization of 5G, Huawei Mate50 has once again become a popular mobile phone with its advanced satellite communication.
Apple subsequently released the iPhone 14 Pro series, which also has satellite functions, confirming the importance that the world’s top manufacturer attaches to the satellite communication functions of mobile phones. Meanwhile, the Android 14 developer preview may launch in February 2023. According to industry analysis, Google may introduce support for satellite connection functions through Android 14 major version updates.
In the semiconductor industry chain, China has been able to mass-produce satellite communication chips that support Beidou’s short messages. Apple’s satellite communication function chips mainly rely on Qualcomm, and mass production can also be achieved.
Therefore, the two factors of software and hardware of the satellite communication function of the mobile phone are all available. With the launch of Android 14, Android mobile phone manufacturers may begin to fully popularize satellite communication functions in the second half of 2023.
At present, the satellite communication function on mobile phones is mainly limited to SOS for help. Huawei uses Beidou satellite communication to increase the trajectory function, which is realized by converting digital short message characters into trajectory coordinates. Similar to the low-speed GPRS and EDGE launched in the mobile communication 2G era, mobile satellite communication Internet access may also follow this path.
5G will develop to the advanced stage
Although 5G is playing a role in vertical industry applications for non-individual users, the general public’s perception of 5G is still not strong enough at this stage. This is related to the fact that the 5G development stage is still relatively early.
Release 18, established in December 2021, is a big step forward in the development of 5G technology and is the first release version of 5G Advanced. R18 content covers XR (extended reality) and media servers, edge computing, AI-based services, network automation, and more. It is expected that the third phase of R18 will be frozen in 2023.
In general, 5G Advanced has three major features: AI, aggregation, and services. AI is used to provide intelligent technical support at all levels during the continuous growth of network scale; aggregation refers to the aggregation of different access modes and networks, including networks of various industries. Services refer to the application of 5G in all walks of life. 5G has evolved from infrastructure to enabling services.
Silicon photonics chips are becoming a new focus of “More than Moore
The chip process in the post-Moore era is mainly divided into three major routes. One is to continue to optimize the CMOS process “Continuing Moore” route, the other is to use advanced packaging technology to realize the “Extended Moore” route of system-level packaging, and the third is to achieve through new materials and new devices “More than Moore” route.
In terms of applications, the value of silicon photonics chips can be well reflected in data center-related scenarios such as large-scale distributed computing, big data analysis, and cloud-native applications, as well as computing scenarios that require extremely high computing power, such as artificial intelligence.
Moreover, compared with ordinary optical modules, silicon optical modules have the advantages of low power consumption, high integration and high speed, which can significantly reduce the cost of data centers that require a large number of optical modules.
From the perspective of the progress of the industrial chain, the global silicon photonics industry chain has gradually matured, and American companies have occupied the leading position in the shipment of silicon photonics chips and modules.
Silicon-based optoelectronic heterogeneous integration technology that uses different materials to take advantage of their respective optoelectronic characteristics is a trend in recent years. At present, the manufacturing platform based on SOI (Silicon on Insulator) substrate is mainly used, and the monolithic integration of detector and modulator has been realized.
However, the optoelectronic properties of the silicon material itself, the accuracy of optical calculations, and the design and packaging of silicon photonic chips have not yet been standardized and scaled up, and still need to be resolved.
