ASMC 2021
Emerging trends in global semiconductor market
Robert Maire, President, Semiconductor Advisors, presented the emerging trends in the global semiconductor market at the concluding session of ASMC 2021.
He started with an update on the semiconductor industry. Chip shortages are happening. There are more ordering issues, than production capacity problems. Automotive makers did not manage the chip supply properly. Trailing edge capacity is somewhat limited. Auto manufacturers also buy cheap chips at low margins. Covid-19 swung demand far enough that affected resiliency. Recovery will probably take over a year.
TSMC model to beat
For Intel 2.0, foundry is all about customer service and process management. Intel has to catch up with TSMC first. And, TSMC is not likely to stumble. There is a minimum 5-10 year project. Intel financials were also hurt by triple extra spending — using TSMC foundry, fix broken issues, and become a foundry. Execution is not guaranteed, and talent is still missing. Intel 2.0 is going to be quite a work, going forward.
TSMC is the model to beat! They are greater than half of foundry and EUV. Their leadership also allows pricing power. That can control Apple, AMD, and Intel, as they are captive. It makes Taiwan more attractive to China. US effort is not significant. Intel has not done well in foundry in the past, and needs to do much better, as well as GlobalFoundries. Adequate capacity also exists.
China’s semiconductor spend also continues. It is hampered by restrictions. The ASML EUV blockade of SMIC has been effective. Memory continues to spend for DRAM and NAND. China is likely five years behind at 14nm (more or less). Recent defaults by Tsinghua calls support into question. Semiconductors is not as easy as solar and LED industries. Continuing to fund the industry is critical, as is government support.
M&A may go away. Large M&A may be dead. China is likely to object as retaliation to restrictions. UK is protecting national interests. Consolidation has concentrated power into very few hands. Committee on Foreign Investment in the United States (CFIUS) and other restrictions may bar China from acquisitions.
Memory has surprisingly remained stable. NAND and DRAM supply has remained relatively balanced. China NAND supply has not on line in a big way. NAND 3D stacking is growing at a sustainable pace. DRAM investments have been on the light side. No excessive investments are happening, as in the past. Bit growth has been steady.
Moore’s Law and EUV
Traditional Moore’s Law geometric scaling has become more difficult. Scaling through packaging helps, such as 2D/3D. Mixed die in packages helps optimize the fab resources. Eventual move to 3D logic designs will resume density drive. Not everyone has made the jump to EUV. Moore’s Law is getting better on packaging.
Transition to EUV boosted TSMC and ASML as clear leaders. TSMC accounts for more than half of all EUV. Intel is lagging well behind. There does not seem to be enough EUV tools going around. High NA and second-gen EUV will increase productivity. The pendulum has swung back to litho taking the lead in Moore’s Law progress. Several generation of tricks are still left.
The Nvidia and ARM deal is not going to happen. The deal is under review by UK authorities. It concentrates huge power in Nvidia. China will likely object if the UK eventually allows this. Blockage is good for AMD and Intel, who are most impacted. ARM has been an impartial supplier of CPUs. ARM is also used in the smartphone industry extensively.
As for investment, valuations are high as multiples have inflated. Scarcity makes the heart grow fonder. Semiconductor equipment is the mutual fund of chip makers. Intel is in a tough spot, and needs to spend a lot. X86 has been slowing down. TSMC remains the company to beat. Nvidia will have a strong play on AI and future technologies. They are well placed. For Micron, Samsung, SK Hynix, etc., memory remains steady, which is good for chips. 2021 looks like a great year for semiconductor investment.
Fab equipment spending to witness record growth: SEMI
Christian Dieseldorff, Senior Principal Analyst, SEMI, presented semiconductor manufacturing market trends on day 3 of the ongoing ASMC 2021, organized by SEMI.
We see three consecutive years (2020, 2021, and 2022) of continuous growth for the semiconductor industry, with each year setting a new record. After negative growth in 2019, 2020 had started optimistically. We expect the milestone of $500 million being crossed for semiconductors.
Companies have increased their capex to record levels for 2021, such as TSMC, Samsung, Intel, etc. Worldwide, will be record high, with $107 billion in 2020, to $140 billion in 2021, on to $142 billion for 2022. Huge increases in capex and demand for equipment may cause a bottleneck for equipment deliveries (increase in tool lead times).
Revenue for the device manufacturers reached record highs, with $107 billion in 2020, and $140 billion in 2021. Fab equipment spending is also going to witness record growth. As of May 2021, it was 30 percent, and for 2022, it will be 14 percent. By region, Korea and Taiwan are leading, with 29 and 27 billion, respectively. China is next with $15 billion. Europe and Mideast will spend $7,6 billion, and USA will spend $6 billion.
Memory will increase by 62 percent, micro by 16 percent, and discrete by 240 percent. Foundry/logic investment is also increasing. For 3nm in 2021 through 2024, there will be 30 percent share. DRAM is mostly at 14-16nm for 2021. Some are using EUV for DRAM, such as SK Hynix and Samsung. 3D NAND is at 19-20nm, and some around 30nm. 3D NAND uses different layers.
Fab equipment spending by wafer size shows 300mm at 93 percent, with 200mm at 5 percent. Wafer capacity by wafer size is growing 6 percent this year, and 6 percent next year. Share of capacity for 300mm is 55 percent, and 25 percent for 200mm. Foundries represent over 50 percent of all 200mm capacity. 200mm foundry capacity is in short supply with demand surging for PMIC, DDI (display drivers), MCUs, and sensors. Foundry capacity is limited, and ramp remains in lower single digits in 2021 and 2022.
Fab growth ahead
Looking at share of semiconductor capacity by regions, Americas dropped from 16 percent to 12 percent. Japan dropped from 25 percent to 18 percent, and Europe dropped from 11 percent to 8 percent. China increased from 8 percent to 19 percent, and Korea and Taiwan increased their share. China is the fastest growing region. It will be a region with most capacity by 2022. Looking at only IC capacity (excludes discrete, MEMS and sensors, and opto), China will be no. 3, below Korea and Japan. Chinese-owned companies are slowly building their share.
Looking at fab counts of operational semiconductor fabs and new fabs, 300mm will add 163 fabs by 2022, while 218 fabs will be added for 200mm by 2022. The count will be higher if we include low-probability fabs. Fab counts by region in 2022 will see 135 new fabs come up in Taiwan, 235 in China, 183 in Japan, 215 in Americas, 165 in Europe and Mideast, 51 in Korea, and 46 in SE Asia, by 2022. Only Intel is US-owned, and others are non-US entities building fabs in the USA.
We see good revenue expectation for 2021 and 2022, with semiconductor revenue reaching the $500 billion mark in 2021. There will be three years with record numbers for capex and fab equipment spending. Huge increases in capex and demand for equipment may cause bottlenecks. Strong increases of investments for foundry and memory sectors are driven by 5G, AI/HPC, automotive, and IoT.
GlobalFoundries on impact of pandemic on semiconductor industry
ASMC 2021 day 2 began with a keynote: A reflection on the impact of the pandemic on the semiconductor industry, delivered by Dr. John Pellerin, VP and Chief Technologist, GlobalFoundries.
Looking back on 2020 Covid-19 has been a catalyst regarding megatrends. Foundries are changing an industry that is changing the world. The global GDP is $91 trillion, of which electronics is about $2.3 trillion, semiconductors is $563 billion, including memory, and $86 billion is for foundry, including IDM foundry.
How was 2020 like?
Some predictions of 2020 for semiconductors were quite bleak. There was a foreshadowing of possible unusual dynamics. Some shifts in work processes and consumer behavior that arouse during the pandemic might persist, and open new markets and routes to markets. Looking back reflected a different outcome for semiconductors overall. The estimated 2020 to 2021 growth was 11-12 percent or $504 billion.
Covid-19 and stay-at-home economy significantly increased chip demand. Six out of eight segments beat the pre-Covid-19 estimated. Wireless, PCs, storage, GPUs and peripherals, consumer electronics, servers, and wired beat expectations. Only industrial and automotive failed to beat the estimates. In the auto chip market, the pandemic created unprecedented shifts in global semiconductor demand. There were sudden and significant decrease in semiconductor auto sales in Q2-2020. This was followed by an equally steep and sudden rise in H2-2020. The deep trough didn’t last long.
2020 was like living a decade in a year. Technology enabled a better normal. The 5G super cycle was here. There were changes in global trade, creating tailwinds in the global semiconductor industry. There was accelerated global demand. Over a billion people have worldwide 5G coverage. There was over 50 percent mobile network data traffic increase. There were 200+ billion hours of additional video traffic. Covid-19 was a catalyst for these astounding statistics.
In 2020, there were record production output and wafer starts. GlobalFoundries was also listed among America’s safest companies. We had to put in geography-specific crisis management teams and protocols. There were Covid-19-specific employee leave programs, and worker arrangements for closed borders. There was continuity and growth of operations amidst extraordinary and difficult global circumstances. 2020 was a strong year from a business perspective.
Lessons from 2020 pandemic playing forward to the strong 2021 demand cycle, included supply fragility. PPE shortages started in January. There was big push for IT equipment through all channels. Safe stock levels of parts and raw materials were challenged. Logistics disruptions saw border closures impacting people and goods moving. Air cargo running on passenger flights were impacted. GlobalFoundries was deemed essential, and extended to 30+ suppliers globally.
There was semiconductor demand surge and permanent demand shift in 2021. In 2021, we began with supply fragility, with diversity and redundancy, not just capacity. There were wafer starts and packaging laminates, etc., under pressure, due to deeper and longer supply agreements, and having coverage over the optimization and efficiency. For capex lead times, there was the equipment productivity improvement of the existing assets. Tom Weber, SVP, said ‘no wafer start left behind.’
Megatrends post 2020
Covid-19 has served as a catalyst. The megatrends accelerating after 2020 include frictionless networking, virtualization, and hierarchical AI. For frictionless networking, the last hop is always RF/wireless. Core is always optical. There is real-time (edge) remote processing, and everything is part of a seamless network.
Semiconductor requirements include RF PA, switch and LNA performance, integration of software/LNA/PA, integrated photonic ICs, and low power RF with embedded memory. Key semiconductor innovations include photonic SoC integration, efficient, high power PA, THz SiGe and SoI integration, and wide band gap.
For virtualization, there are brontobytes of data manipulation, memory and compute converging, photonic and quantum compute, and disaggregated storage solutions as market implications. Semiconductor requirements include ultra-low power computing, imaging, and display, and data transport, open source scalable secure protocols, co-packaged heterogenous systems, etc. Key semiconductor innovations include optical compute, co-synthesized logic and memory, 2.5D/3D hetero-interconnect, and THz SiGe/SoI/CS.
For hierarchical AI, market implications include voice commands and processing at the edge, AI embedded in devices, and explosion of edge devices (industrial IoT). Semiconductor requirements include security, reliability and failover management, and speed, especially, shorter response times. Key semiconductor innovations include platform for efficient and comprehensive integration.
DCs, auto, 5G/6G
These megatrends apply to specific segments, such as data centers, automotive, and 5G/6G. Data centers are on a power collision course. By 2040, computing will use 100 percent of world’s energy production. There are missions ongoing to flatten the power consumption curve, with novel compute architectures, optical networking, and AI everywhere.
Megatrends are driving innovation in the data center. A disaggregated approach will see more of optical networking, low-power connectivity, efficient power delivery, novel compute architectures, and higher utilization and reduced power, leading to lowered TCO.
There are silicon photonics solutions in the data center. There are pluggable modules for DCs and co-packaged optics or DC interconnects. SiPh today, is from 10km to metro DWDM (400G), upto 2km (400G), and 100-500m (40G-100G). Cu today is transitioning to SiPh. SiPh is becoming the key enabler for data centers.
In computing, novel compute architectures are delivering More than Moore. They are delivering high performance without scaling. There are paradigm shifts with photonic super computing and quantum photonic computing.
GlobalFoundries solutions for moving, connecting, and storing data from edge to cloud, include silicon photonics solutions, SiGe BiCMOS for high-speed interconnectivity interface, and key technologies facilitating the disaggregation of computing, storage, and data transfer to reduce capex and opex.
For automotives, key trends include connectivity, electrification, and safety and automation. New vehicle architecture includes next-gen zonal architecture such as smart sensors, advanced compute and AI, and new infrastructure for power and connectivity. Each zone is responsible for each system.
Smart sensors include ADAS radar, and ADAS LiDAR. In compute and AI, there are advanced AI systems, zone controls, and infotainment. For infrastructure, there is battery management and EV drive control, safe/secure wireless data, and high-speed in-vehicle data. There should be a closer relationship between the foundry and the auto industry. This will lead to direct working exchange, engineering collaboration, and strategic planning and investments. We can drive supply chain to the technology of choice.
For 5G and 6G, we are in the early stages of 5G. 6G will arrive after 10 years. For 6G frequencies, over 100GHz means that technology development must start early. We will use mmWave for 5G and 6G. Huge amounts of spectrum is available for 100-300GHz. This is an enabler for communications and sensing. Large bandwidth and smaller wavelength will be used. W and D bands are the leading contenders. 6G will also drive new RF technology.
For 5G, GlobalFoundries has solutions such as 45RFSoI and 22FDX. There is unparalled 5G mmWave FEM performance. For 6G, silicon is capable at D band ~15dBm, but PAE is low. In InP, best Pout and PAE, but lacking high-volume commercial availability. Scaled GaN has highest power density and high Pout at ~25dBm. As for SiGe, there is a 600GHz roadmap for significant improvement in high-frequency performance. Putting 6G in context, it will be 10-100x better over 5G in figures of merit.
Looking at the future of communication, future networks will be driven by intelligent edge nodes based on high speed, always-on devices. Technology innovations are needed to leverage a vast, untapped spectrum (100GHz-1THz). Innovative RFSOI, FinFET, and SoI/SiGe-based photonics are the platforms for the future.
In summary, the semiconductor industry continues to show leadership, adaptability, and resilience in times of this crisis. Pent-up demand is compounding upon an already growing base. The megatrends intersect the critical segments of the global economy driven by semiconductors: data centers, automotive, and 5G/6G. A diversity of semiconductor solutions are critical for driving the global economy.
Advanced manufacturing with innovation, education and public-private partnership
There was a workforce development panel discussion on Advanced manufacturing: Success demands innovation, education and public-private partnership, at the ongoing ASMC 2021.
The panelists were: Joel Hartmann, Executive VP, STMicroelectronics, Ms. Santosh Kurinec, Prof. Fellow IEEE, Rochester Institute of Technology, Philip Wong, Chief Scientist, TSMC, Ms. Amanda Scarnati, Research Analyst, Citi, and LaMar Hill, Office of President, NY Creates.
Joel Hartmann, ST, said there is an European partnership program. The Grenoble, France, example is of a complete microelectronics value chain. ST products are integrated into national education programs. There are events to engage girls in science and engineering, and activities in primary schools. ST networks with companies.
LaMar Hill, NY Creates, added they have ready sites that are fully enabled, zoned and permitted. You will need 3-5 years and an investment of $200-400 million per site. There is a global R&D backbone and regional resources. New York site has been doing collaborative research for the last 30+ years. Primary stakeholders include New York State, US Government, industry, and Suny RF. Workforce development is very critical. It partners nationally with SEMI to enable SEMI Works. There are enhanced training facilities for things like mechatronics, thin-film processing, and characterization/metrology.
Ms. Santosh Kurinec, Rochester, said they are sustaining the pipeline for semiconductors. There is huge industry demand. They need government and industry support for scholarships, university engagements, etc. Domestic workforce is currently weak, and needs to be improved. We should incentivize students through merit-based scholarships. We should support UG educational programs, and provide industry/national labs based on co-op opportunities for students. They also collaborate with community colleges for developing advanced manufacturing curriculum, and scholarships, etc. We are maintaining a high-quality microelectronics lab and state-of-the-art curriculum. We need government and industry support to sustain.
Ms. Amanda Scarnati, Citi, said the government incentives change the semiconductor landscape. A new advanced logic or memory fab costs ~$20 billion. US fabs are more expensive than Asian fabs due to lower government subsidies. National security is a key driver of wanting fabs within geographic boundaries. NDAA and US Chips Act are some incentives to promote the semiconductor manufacturing. EU and Japan are also looking at expanding the semiconductor capacity. The Biden administration has given $50 billion for developing the industry. An increase in fabs and capacity will drive equipment spend to $100 billion.
Philip Wong, TSMC, noted that TSMC will invest in advanced technology R&D to ensure continued success. The future holds many more possibilities than before. Innovation and clear-eyed vision is necessary for success. We need to develop sustainable, top quality talent pool for TSMC. It supports university research for student training. SRC, USA, U-Tokyo Alliance (Japan), Purdue University (Secure Microelectronics Center), and major industrial affiliate programs such as MIT, Berkeley, Stanford, UCLA, Georgia Tech, etc. There are one-to-one joint development projects (JDP). Several industries are global, and this can be replicated. Many companies are higher in the value chain, and are hiring.
Spotlight on talent
Does talent have to be at the same level as a PhD? Ms. Santosh Kurinec said the industry has been generally hiring PhDs, and the UGs were getting ignored. They can be added to the CMOS labs. LaMar Hill said labour costs need to be discussed. Workers will seek advanced degrees as they grow.
Joel Hartmann said funding from governments is needed. We need to have infrastructure in place. PhD is not the only skill that we need. We have a good network of engineering schools in France, so it is not difficult to hire. Ms. Amanda Scarnati added that about 70 percent is related to government funding. Globally, it is about 40 percent. There is a delicate balance.
Is there an economy of scale that can be reached in fabs? Wong said it is alright, if it is not focused at one place. That is not an essential element. It also helps to move engineers around. Talking about interdisciplinary programs, Ms. Santosh Kurinec said they bring it all together. We have interdisciplinary programs, and other universities need to do that. Hill agreed that you need to bring the various disciplines together. Children also need to be pulled into the system. We have to drive the aspects of our own industry into the schools to get more people to join us.
Are we now in an arms race for process development? Everyone wants manufacturing localized. Hartmann said this is not that necessary. Ms. Scarnati said it is more about national security. We have to balance this. We are also talking about the large suppliers.
How long does it take to ramp up a workforce? Hartmann said it depends on the type of people getting hired. They need 3-4 years of training. We have designers, technologists, etc., working together. We also need to have the new manufacturing technologies, and new hires, for example, in materials. We also need to address issues like sustainability and being carbon free, etc.
Ms. Kurinec said that if students go for internships, they can learn much faster. We have students transferring to our program. We must encourage more scholarships. Hill added that you can do that without incurring new costs. Students do not need a 4-year degree to be a technician. Ms. Kurinec added that we need to have semiconductors in the K-12 curriculum. We also have to do something as a society, and an industry. National science scholarships can bring more students into the system.
Ms. Scarnati said there has not been much work on salary differentials. It is more about building the infrastructure and beyond. Hartmann said we need to maintain balance with competitors in our regions.
Women, taxation, location
How do we get more women into the semiconductor industry? Ms. Kurinec said it all starts from K-12. That’s where we have to focus on young girls. They should have role models to look up to. They can be smart engineers. Hill said there are some phenomenal role models for women.
Regarding taxation, Ms. Scarnati said it will impact where fabs will be located. We need to support better tax policy within the USA. Additional regulations are needed. Safety is paramount. Additional investments are also well managed. We need to have better tax regime. Hill added that if you invested $10 million, you would need 8-9 years to recover that. Difference in regions can occur how taxes are treated.
Hartmann said Europe is getting new regulations on safety. It is bit complex. There are some materials and chemicals that can be difficult to qualify. This is forcing us to be more focused on quality and safety. Ms. Scarnati said that some jobs are remote, and are part of critical infrastructure. Technicians and engineers have to be located on site.
Hill said there are companies who have offices near rivers. Hartmann added that you also need to look at the tax incentives, besides looking at the network of people, education, etc., when you put up a fab. Ms. Kurinec said that we should also give some incentives to students. An educational roadmap must be drawn for semiconductors.
Ed Sperling, Editor, Semiconductor Engineering, was the moderator.
ASMC 2021 begins with continuing Moore’s Law!
Advanced Semiconductor Manufacturing Conference (ASMC) 2021, organized by SEMI, USA, commenced today.
Dave Anderson, President, SEMI Americas, welcomed the audience. AMCS continues to be the leading conference for semiconductor manufacturing. This includes smart manufacturing, AI/ML, etc. There are plenty of challenges yet to solve. ASMC continues to increase in focus and value for the chip manufacturing community. We are witnessing unprecedented investment in fab capacity in the USA, and increase in global manufacturing. He welcomed everyone to attend the Semicon West 2021, to be held in December 2021.
ASMC 2021 co-chairs and conference moderators, Ms. Alexa Greer, Technical Program Manager, AI, Software & Physics Modeling, KLA, and Ishtiaq Ahsan, Senior Manager, IBM, welcomed the participants. Ms. Greer said there is a great line-up of keynotes, presentations, etc.
Continuing Moore’s Law
The opening keynote on Continuing Moore’s Law was delivered by Dr. Gary Patton, Corporate VP, GM Design Enablement, Intel. Thirty years ago, technology was largely for the rich. Today, it is for all. Every industry will be digitally remastered over the next few years.
Moore’s Law is all about economics and increased economic efficiency. Scaling and elasticity enables more development. Transistor scaling drove product performance. Scaling has been driven by the integration of new architectures and materials, and limited by precision of equipment and metrology. We also replaced silicon dioxide. In silicon scaling, major technology innovations saturate after about a decade. Disruptive innovations will enable the next decades of progress. We went to 3D devices, such as FinFETs, etc. We got to an atomic dimension limit by 2020.
Data is now everywhere, and increasing exponentially. It is expected to reach 160ZB of data by 2025. AI is everywhere to process high volume inbound data. xPU diversity is addressing the diverse set of unstructured data. Compute embedded is throughout the cloud-network-edge to manage QoS and security.
Customers value information. We can convert unstructured data into services and information creates enormous value. Hardware solutions are necessary, but are insufficient. Compute everywhere in the network is embedded. Compute is chasing the data. It is increasing the intelligence at the edge. Distributed computing drives domain-specific architectures. There are custom silicon apps.
Computer barriers include complexity and cost. DARPA called it the curse of Moore’s Law. There is increasing cost of hardware design and verification. There has been continuing evolution of Moore’s Law. It means scaling architectures and materials. There will be heterogenous system integration. FinFET moved to nanowires or nanoribbons for improved electrostatics. There are novel design and architectures for scaling. Density scaling includes gate pitch scaling and cell height scaling.
We are doing research on 2D material nanosheet transistor. Tungsten disulfide is a promising 2D channel material. There is monolithic 3D transistor stacking, with germanium PMOS and silicon NMOS devices. There are new apps with monolithic GaN NMOS and Si PMOS, single chip fully integrated 5G RF FE, etc. Intel is pursuing monolithic memory stacking. Interconnects also create complex interconnected architectures. Novel architectures power delivery and routing options. There are some exciting technology options for continuing Moore’s Law.
3D solutions lead to system optimization. Heterogenous integration improves performance, etc. There is also heterogenous integration with advanced packaging and interoperable tiles. There is continued scaling, improved performance, etc. Package interconnect trends include improved power efficiency as bump density scales.
Intel’s embedded multi-die interconnect bridge (EMIB) is a small bridge die with multiple routing layers, and embedded only under the edges if two interconnecting dies. There is Foveros technology, with advanced 3D face-to-face die stacking packaging. This helps to interconnect density scaling and lower wire parasitics. EMIB and Foveros were designed to complement each other. Hybrid bonding interconnect will continue densification. Area scales with bump pitch. There is also the package as a platform for innovation. Intel is shipping Agilex FPGA 2.5D and Lakefield 3D.
DTCO or design and technology co-optimization is increasingly important. This is a fundamental element for Moore’s Law scaling. Some practical examples showed process was co-optimized with standard cell and IP block layout is continuing area scaling, and improving performance and power efficiency. DTCO can be improved using AI. AI/ML methods deliver superior results.
System technology co-optimization or STCO is where DTCO has moved. It drives system-level decisions. We can do some things with DTCO. We can modularize and re-factor. We can also do layer and abstract. We can also automate and scale. Hardware optimization results in faster runtimes. We will however face future verification challenges. Sequential verification will no longer work. Here, we will require parallelization. Moore’s Law is still alive. We will continue to innovate.
PRADEEP's TECHPOINTS 