Memory markets struggled through a mixed 2020, but appear to have emerged, strongly positioned. Can 2021 be the beginning of another memory super-cycle?
Walt Coon, VP of NAND and Memory Research, Yole Développement, Semiconductor, Memory & Computing Division, said NAND was in early stages of recovery exiting 2019. A record year was expected in 2020. Profitability during the downturn was high. DRAM was in downward trajectory. Recovery was expected to commence in 2020. The pandemic led to H1-2020 demand pull-in. It was a weaker H2 for both NAND and DRAM. Work- and learn-from-home transitions led to strong notebook/Chromebook usage. Covid-19 impacted memory profits. Memory markets exited in weaker position during 2020.
Speaking about the market outlook for 2021, Mike Howard, VP of DRAM and Memory Research, Yole Développement, Semiconductor, Memory & Computing Division, said there was significant capex cut off in 2020 for DRAM and NAND. This is expected to grow 20 percent in 2021, and 2022, respectively. Suppliers also cannot pull back on their capex as other suppliers may come into play.
DRAM bit density is expected to grow 13-15 percent. Similar for NAND, growing at 20 percent. DRAM production is forecast to grow 18 percent in 2021, and 19 percent in 2022 based on recent capex, and expected growth in bit density per wafer. Significant adjustments could be made in 2022 production growth, and could reach over 22 percent. NAND production is forecast to grow 32 percent in 2021, and 31 percent in 2022, based on recent capex and growth in bit density per wafer.
On the demand side, server shipments are likely to grow inline with 5-year average in 2021, and slightly below in 2022. DRAM content is expected to grow inline with the recent trends. NAND content growth is expected to decelerate from 2020 growth rates, but remain above the 2019 rate.
Smartphone unit shipments are likely to outgrow recent trends. It will grow 6 percent in 2021 and 3 percent in 2022. Content growth is likely to continue the deceleration of the past few years. PC shipments are likely to remain high in 2021, before falling off steeply in 2022. Content growth is expected to decelerate slowly for both NAND and DRAM.
2021 is expected to see rather high DRAM demand growth and NAND demand growth slightly above the trendline. It will be a good year for memory overall. Demand pull-ins will happen in both markets. Prices will climb through 2021, peaking in Q4-21 for NAND and Q1-22 for DRAM. NAND revenue will peak or plateau in Q4-21. DRAM revenue will peak in mid-2022. The overall memory revenue in 2021 is expected to hit $163 billion and rise to $196 billion in 2022.
Seok-Hee Lee, President and CEO of SK hynix, delivered the opening keynote at the ongoing IRPS 2021. He talked about memory’s journey towards the future information and communications technology world.
Lee highlighted on some changes in the ICT environment, followed by new drivers of the memory industry, and collaboration under digital transformation. He concluded with future direction of the semiconductor industry.
Comparing 2019 and 2020, the impact of Covid-19 has seen IoT devices subscriptions rise. There is also a rise in the number of AI speakers sold, 5G data usage, and the number of public open data. It has also accelerated the digital transformation of Industry 4.0. 5G is a game changer for the world. It will create new values for numerous industries, such as AI, autonomous vehicles, smart homes, AR/VR, etc. These changes are sending ripples across the entire ICT industry.
Autonomous vehicle/car is yet another game changer. Vehicles, powered by connectivity, are becoming more than just a means of transportation. A single autonomous car is expected to generate 4,000GB data each day. It has been a subject of several movies envisioning the future. We hope to see this happening in the ICT world. The car should be able to receive, analyze, convert, and feedback the data from the various sensors, such as cameras, LiDAR, etc.
There is also an explosive growth of structured and unstructured data. With 5G and autonomous cars, this is expected to increase in a real-time basis. the amount of structured data created and stored in the various electronic devices that we use increases. The amount of unstructured data is expected to increase, as well. The unpredictable, unstructured data is created and consumed in M2M communication, including V2V communication. The total amount of structured and unstructured data is expected to increase exponentially.
To handle this explosive growth of data, the CAGR of data centers around the world is expected to reach 15 percent. The data center market is expected to double by 2025. The data center market is also expected to double by 2025. If you look at the capacity of DRAM and NAND flash for each data center, the numbers are daunting. The demand for memory and storage essential for data centers will inevitably increase. The semiconductor memory will become more important in the future.
With digital transformation, the role of memory will further expand. The function of memory itself is becoming more diverse. The role of semiconductor memory will indeed, be huge. There are memory requirements and reliability. There are three key things: high speed, hyper scale storage, and ultra-low power. Along with this, there is the need for high reliability. All of this is needed for things such as autonomous cars and smart healthcare. Here, the components are expected to run for a long period of time under the extreme conditions. Reliability will also become one of the major requirements.
It is time for the semiconductor industry to respond to the environmental issues related to energy. In the era of digital transformation, there are 3Ss that are critical. Scaling the current and future scaling sustainability, memory industry’s social responsibility, and smart changes in the role of memory to respond to the demands.
Look at the scaling value, we have done a lot to overcome the limitations of scaling. In DRAM, the trend started from 20nm class in 2012 to 10nm class in 2017, to meet the industry required speed and density. The performance also improved from DDR4 to DDR5. DDR5 is also the next-gen DRAM specification for high-speed, high-density products optimized for Big Data, AI, and ML. The data transfer rate for the first-generation DDR5 DRAM reached 4800-5600Mbps, which is 1.8x faster than DDR4. In the near future, the maximum transfer rate of DDR5 will be over 6400Mbps.
The error correction code or ECC is embedded inside the DRAM chip. This helps self-correct errors down to a single bit level in DRAM cells. We believe the reliability of the system will improve by approximately 20 times. SK hynix also contributed by developing and launching DDR5 for the first time in the industry.
There are challenges in DRAM scaling, such as lithography, cell capacitor, and low resistance. In lithography, there is fine pattern with simpler process. For cell capacitor, we have high-K di-electric material innovation. For low resistance, there is new metal with lower resistance. We are focused in three areas. to overcome the limitation of patterning in lithography, maintaining the capacitance of the cell capacitor, and lowering the resistance of the metal for the interconnect. The last one is not a new item. In logic technology, it has been an issue for the last 10 years. Now, it is getting into memory.
To overcome the limitation of lithography, we adopted EUV technology to secure a stable EUV patterning performance and enhance productivity. We are also controlling the materials and defects, in addition to developing photo resist materials, by working with partners.
With regard to the cell capacitor, there has been innovation in new material and structure to maintain cell capacitance. It becomes smaller as di-electric thickness scaling continues. The amount of charge stored in the capacitor is the key for a successful DRAM device. For low resistance, we have a next-generation electrode and insulating materials, and new processes.
Soft error provides a challenge to DRAM reliability. Especially, for automotive app, responding to the soft error is very important. To increase immunity against soft error, design option is usually considered. SK hynix overcame this challenge with extreme engineering, such as well structure and junction size optimization, in order to achieve soft error rate immunity and cost objectives.
NAND flash challenges
Next, we come to NAND flash. From scaling to stacking in 2015, and after transitioning from 2D and 3D structure in 2015, we are securing 4D NAND technology in 2018. NAND development continues with the goal of achieving high capacity by overcoming the limitation of stacking.
In 2020, SK hynix successfully developed the industry’s most multi-layered 176-layer 4D NAND. In NAND, there are three key challenges we continue to face. Those are securing high aspect ratio contact etch technology, securing cell di-electric characteristics, and solving film stress issues. We came up with 4D NAND stacking to continue to achieve the industry required high density.
To further enable stacking, high aspect ratio contact (HARC) technology is required. With partners, we plan to improve the equipment performance, and come up with innovative materials. We will also innovate the structures of HARC etch technology for more stacking. We also need to secure the cell di-electric characteristics, for which, we are using atomic layer deposition (ALD) technology. It allows fast and uniform deposition of different types of materials in a small plug.
We are also trapping and maintaining a certain amount of electric charge.
Lasty, we are also looking to minimize the film stress from causing wafer bonding and warpage challenge of NAND stacking. The current solution used is to control the mechanical stress levels of films, and optimize the cell oxide and nitride materials.
There is also the NAND reliability challenge in oxide nitride (ON) scaling. There is large bake retention due to the lateral charge spreading. Charge trapped layer engineering and isolated CTN structure are used to solve the issue. There is the lateral charge loss in ON pitch scaling. SK hynix has developed a deep trap CTN (charge trap nitride).
Roadmap for next decade
Let’s look at how the DRAM scaling will continue below 10nm and NAND stacking continues beyond 600 layers. In about 10 years from now, we can expect a challenge requiring a change in the technology to continue scaling DRAM below 10nm, and NAND stacking beyond 600 layers. SK hynix will overcome those challenges with innovations in materials and structure. We will bring innovative technology to further scaling in DRAM, and stacking in NAND.
Addressing social needs
Recent global climate and temperature change is accelerating. A 1 degree C change in average temperature brings more frequent killer heat waves. Countries like US, China, and even Europe, have declared net zero targets to reduce CO2 emissions. The EU has pledged to cut greenhouse gas emissions to net zero by 2050. The energy sector faces increasing pressure to curb CO2 emissions.
SK hynix will innovate in NAND. It also plans to reduce all HDDs with low-power SSDs, so that CO2 generation can be reduced by 93 percent. Imagine what it would be to replace HDDs of all the data centers around the world with low-power PLCs or QLC-based SSDs by 2030. PLC stands for penta-level cells, which means you can put upto five levels of information in one cell. We can also create approximately $3.8 billion worth of social value.
There is demand for low-power products that are capable of transferring massive workloads at higher data transfer rates than before. The speed of memory is yet to catch up with the demand. With the explosive growth of data, the existing memory is not enough. Changes in computing/memory architecture are in progress to handle increasing data/computation volumes. The architecture of the CPU is evolving into multi-core and many-core, and heterogenous computing. There is also a demand for heterogenous memory. There are some representative examples of heterogenous memory solutions.
One is, high-bandwidth memory. It is optimized for AI and HPC. Next, comes the ultra-low-power memory. It is high-density memory compared to cache, and has significantly better power efficiency compared to the existing memory. Last is the emerging interface solution, the CXL, or the compute express link. CXL memory is being prepared as a solution. It realizes the value of persistent memory. SK hynix has continued to develop various solutions, and will continue to do so in the future. We can also enable total computing energy saving by smart solution. Convergence solutions for eras need to be delivered.
New drivers for memory
The future world will be hyper connected, characterized by low power consumption. The future will be the era of new ICT, with AI-based everything integration. All the devices will be connected, and integrated with AI. The value that reflects this future world is ‘smart’. We will continue to do our best for memory transformation for new ICT era. We will strive to incorporate smart value into intelligent memory solutions.
The next 10 years will see the memories get converged with CPU and storage. Revolutionary technology, such as neuromorphic computing will be required. There will also be a convergence of memory and logic. The concept is to add a few computing functions of the CPU to the DRAM. As the speed was increased in high-bandwidth memory (HBM), it will increase further in processing near memory (PNM). The future will be process in memory (PIM), and ultimately, computing in memory (CIM). The CPU and memory are integrated within a single die to deliver high-performance computing system.
Memory solution in the integration era will be a simultaneous execution of computing and storage. There will be neuromorphic memory, and ultra high-speed computation and storage will become possible. It will be possible to enable a true, smart world.
SK hynix is preparing for the memory journey in the future, and moving on to the neuromorphic era. The revolutionary path is in the post-Von Neumann era, with neuromorphic, DNA storage, etc.
Semiconductor industry needs transformation
The semiconductor industry also needs to transform itself. Over time, it has been capital- and knowledge-intensive in nature, to overcome the various challenges. Now, It is becoming increasingly difficult to develop technologies. There are also challenges regarding scaling sub-10nm class, dealing with the increased number of process steps, and over two months of prolonged turnaround times.
Transformation is now needed in the semiconductor industry. We have to deliver social and smart value. Various companies in the semiconductor industry should collaborate with one another. Open innovation, based on collaboration, and sharing with customers, suppliers, academia, and governments, can help us all shape a new era. SK hynix will take the initiative in enhancing the quality of our lives. We will incorporate scaling, social, and smart values.
Here is part two of the SEMI Silicon Valley Chapter and SEMI Northeast Chapter conference on semiconductor outlook for 2021.
Dan Hutcheson, CEO, VLSI Research, presented on how the pandemic has turbocharged digitization. There are macro factors driving the semiconductor industry. Covid-19 closed one door and opened another. We have since gone from rainy situation to a sunny situation. The semiconductor industry had prepared the world for Covid-19. This year has started really strong. IC market growth has been slowing because of Covid-19.
This year, VLSI Research’s forecast for semiconductor equipment industry is about 20 percent. Zooming in on critical IC market segments, semiconductor sales growth should be +13 percent.
Focusing on IC market growth this week, the 13-week MA shows 2021 has kicked off on a strong growth path. Unlike the last 2 years, DRAM, NAND, auto, analog logic, and power are in a tight growth pattern – most recently ranging from +14 percent to +19 percent. In 2019, DRAM and NAND sung the blues after a hot 2018. 2020 was the year for auto ICs to be blue. With all the news of an auto IC shortage, this market is clearly hot. The auto IC sales are forecast to grow 21 percent. Seeds of auto IC shortage in 2020 were sown in poor supply chain management.
DRAM is very data-driven and forecast to grow 21 percent. NAND IC sales is forecast to grow 15 percent, as well. Analog IC and power discrete sales is forecast to grow 12 percent. Logic IC sales is forecast to grow 10 percent.
IC supply/demand trends
Looking at the IC supply/demand trends, IC wafer fab production levels continued to rise last week at ~20 percent above 2020 levels, with production levels above 2M 300mm equivalents per week. Supply/demand status held tight for the week. DRAM jumped, NAND and IDM tightened, while foundry, OSAT, and analog and power loosened. The 1Q nowcast is tight
In semiconductor utilization, all sectors are high. These include wafer fab, test, and packaging. Electronics’ prices are also declining. These include TVs, PCs, notebooks, tablets smartphones, cell phones, digital cameras, appliances, etc.
Looking at semiconductor inventory, the inventory-to-billings ratio is in an expansionary range, and ~0.25 below critical levels. Total IC inventories are in decline, suggesting high fear levels at the start of 2021.
Semiconductor/semiconductor capital equipment 2021 outlook
Harlan Sur, Executive Director, JP Morgan, presented semiconductor/semiconductor capital equipment 2021 outlook and long-term trends. Semiconductors/semicap stocks have outperformed the market over the last 1, 3, 5, 10 years.
Drivers of the strong stock performance include the realization that semiconductors are the foundational building block for all innovation in the technology sector — applications/devices/appliances are getting more intelligent and requiring higher levels of semiconductor $ content. There is lower cyclicality in the industry driven by end-market diversification and disciplined supply growth. Lower cyclicality drives more focus on profitability and free cash flow expansion, leading to strong capital return to shareholders.
Industry consolidation (M&A) is expected to drive diversification, R&D scale, and enhanced profitability and capital returns. Near-term, there is positive Y/Y inflection in semiconductor company revenues in 2H20, and expectations of industry growth through 2021. We expect long-term positive fundamental trends to continue going forward. Semiconductors have been in more stable growth phase. There is focus on market leadership, strong product cycles, margin/free cash flow expansion, and capital allocation.
End-market diversification and lower industry cyclicality drives more focus on profitability and cash flow. As operating margins and free cash flow margins have expanded, companies put in place strong capital return programs, driving higher valuation multiples. Industry consolidation has driven valuation multiples higher over time.
Semiconductors/semicaps growth and cyclical trends remain positive entering 2021 and beyond. We believe the semiconductor industry has entered a more stable and less cyclical growth phase characterized by low- to mid-single-digit annual revenue growth and high-single-digit unit growth. With the industry generally driving high-single-digit Y/Y unit growth, the entire value chain is able to better predict silicon consumption requirements, better respond to perturbations in supply/demand, and more efficiently plan manufacturing output. As a result, volatility in semiconductor supply/demand and semiconductor equipment spending has muted significantly.
Compare this to 15-20 years ago, when unit growth rates were +15 percent Y/Y – small perturbations in supply/demand would drive significant swings in inventory, shipments, capacity planning, and equipment spending. Bottom line: The current environment is likely more stable and less cyclical for semiconductor and semiconductor capital equipment suppliers. In a maturing industry, we believe the market will focus on market leadership/scale, operating margin and free cash flow margin expansion, and increasing payout ratios.
We see the semiconductor industry revenue up 10-12 percent Y/Y (bias upward) in 2021, following a 7 percent Y/Y growth in 2020. If you recall, the 2H20 demand picked up significantly and growth turned positive after 1H20’s weak demand environment and supply chain disruptions (Covid-19). In semiconductor equipment, we see spending up ~16 percent Y/Y in 2021 led by DRAM and foundry strength.
As for supply constraints across all end markets, we see multiple quarters of strength for the semiconductor suppliers. Channel/customer inventories are at/near historic lows, and lead times are continuing to get stretched out. Given the strong demand environment combined with supply tightness, we anticipate strong demand trends through 2021.
We expect continued industry consolidation (M&A). There is focus on scale, diversification, and margin and FCF expansion. There is promising outlook for foundry/memory in 2021, with demand and capex spending driving strong semiconductor equipment fundamentals. Potential investment in US domestic manufacturing capability is a positive, with innovation and assurance of the supply base.
Industry consolidation should also support valuations. We can expect more M&As to happen. Semiconductors are consolidating focus on building scale and driving profitability improvements, End-market diversification, etc. Consolidation should drive more stable revenue growth and improve margins. Less competition leads to less pricing pressure. There will be more market leadership and diversity. In 2015-2018 have seen $100 billion+ per year in M&A deal activity vs. the $20-30 billion run-rate prior to this time period.
M&As are characterized by big getting bigger. Going forward, we expect to start seeing a lot more M&A activity with the smaller/medium-sized companies, as there is pressure to drive scale to compete with much bigger competitors.
Data center fundamentals strong
Strong data center fundamentals, led by cloud service provider (CSP) spending are driving strong demand for compute, networking, and memory/storage semiconductors.
Look for companies levered to data center trends to outperform in 2021 across compute, networking, and storage/memory after digestion cycle in 2H20. CSP spending (top 4) grew by 10 percent in 2020 and was up 6 percent in 2019. We expect cloud spending to reaccelerate in 1H21 and grow 25percent+ in 2021, and at 10-15 percent CAGR over the next few years.
Cloud services revenues continue growing >40 percent + Y/Y. Over the next 5 years, CIOs should grow spending on public cloud by 4x. Early ramp of new processors by Intel, AMD, Nvidia, and ARM will see adopters. Silicon switch ports (>25Gbps) should grow 23 percent CAGR. DRAM memory content in a cloud server is 50 percent higher than traditional enterprise server – OW MU. Data center compute acceleration is growing >25 percent CAGR, driven by higher complexity workloads (AI/Deep Learning, analytics, etc.).
Resurgent custom chips
Custom chip (ASIC) market is experiencing a resurgence in activity as large OEMS, cloud, and hyperscalers look to differentiate at the silicon level — $10-$12 billion silicon market opportunity. Demand is rising for custom ASICs as many large OEMs/CSPs/hyperscalers are looking for more differentiation, better performance, lower power consumption and overall lower cost of ownership versus off-the-shelf chip solutions (or ASSPs).
These same customers do not have the capabilities to do large complex SoC) designs, nor do they have the broad IP portfolio of on-chip design blocks, like high-speed SERDES capabilities or high-speed memory interface technology. They need to engage with semiconductor companies (ASIC companies) that have the IP and chip design expertise (Broadcom, Marvell, Intel, MediaTek as examples).
The digital custom ASIC chip market is a ~$10-$12 billion per year market opportunity. These include cloud/hyperscale ASICS (AI processors, smartNICs, security/video processors, networking/storage acceleration). Telco/service provider equipment OEMs also see growth, for 5G base station modems, 5G digital front ends, 5G MIMO/beamforming DSPs, and coherent DSPs for long haul/metro.
There are 5G opportunities too. These benefit wireless infrastructure and wireless RF smartphone market leaders. 5G base station deployments are growing by more than 22 percent CAGR (2020-2023E). 5G base station estimates are growing from 2020’s ~800k to 2023’s ~1400k. We expect North America activity to pick up in second half of the year followed by Europe in 2022.
Digital semiconductors are growing from ~$3 billion in 2020 to ~$4 billion in 2023. Analog semiconductors will grow from $0.8 billion in 2020 to $1.1-$1.2 billion in 2023. This is a positive for players like Broadcom, Marvell, Intel, Analog Devices, Xilinx, Qorvo, NXP, etc. Massive MIMO/Beamforming are key enablers of 5G sub-6 GHz and mmWave to increase network capacity, data rates with better energy efficiency and TCO. GaN opportunity scales with number of antenna elements. GaN market will grow from ~$350 million in 2020E to ~$550 million in 2023E (15-20 percent CAGR).
5G smartphone complexity benefits the RF market leaders such as Qorvo, Skyworks, Broadcom, etc. The market is growing at 10-12 percent CAGR (2019-2022E). 5G smartphone estimates are growing from 2020’s 225 million to 2022’s 725 million. Ramp of 5G to meaningfully increase RF market opportunity primarily on new sub-6 GHz content, millimeter wave, are also additive over the next few years. There will be ~$5-$7 of incremental 5G sub-6 GHz content. This is positive for Qorvo, Skyworks, Broadcom, etc. Core base component expertise will grow, across PAs, switches, premium filters, etc. It is difficult to insource with lack of foundry model/merchant filter vendors.
Demand growth has been accelerating in memory. Pricing should improve meaningfully in DRAM in early 2021, while price declines in NAND are still moderate. Bit demand in DRAM and NAND should accelerate in 2021. DRAM bit demand should increase to >20 percent with strong demand for server and mobile DRAM. The NAND bit demand should increase to ~40 percent led by SSDs and mobile devices.
There is supply tightness in DRAM, as a result of lower DRAM capex in past two years. It should lead to improved pricing and ASP increases in 2021. NAND market is still in oversupply. ASP declines are set to decelerate later in 2021.
Capital intensity has been increasing across the device types. This is a positive for semiconductor equipment. Increasing capex should drive bit growth for DRAM and NAND. NAND capital intensity for 12X layer should be >50 percent higher than 4X layer NAND. This will require increasing the capex to drive bit growth for DRAM and NAND. Capital intensity is also increasing for foundry/logic, even as EUV has begun ramping. There is 5nm capital intensity that is >50 percent higher than 14nm/16nm. Increasing capital intensity is positive for semiconductor equipment companies, as spending on equipment will likely have a higher floor and be less cyclical over the next several years.
Looking at the wafer fab equipment (WFE) forecast and key programs for semiconductor manufacturers in 2021, we estimate WFE spending is on track to increase by ~16 percent in 2021 to nearly $70 billion.
We expect memory to recover in 2021 led by DRAM on improving supply/demand fundamentals with foundry/logic spending sustainable. Key drivers include following muted memory WFE in 2020 that was held back on supply discipline, we expect memory spending to accelerate to double-digit percent in 2021 led by DRAM. We expect continued foundry/logic spending strength in 2021. This will be broad-based across leading and lagging edge technologies. China spending should remain strong in 2021 as local manufacturers come up the learning curve.
SEMI Silicon Valley Chapter and SEMI Northeast Chapter organized a conference today on Semiconductor Outlook — Navigating Through Turbulent Times – Is the End Near?
David Anderson, President, SEMI Americas, said that the pandemic saw our homes become offices. We had to adapt quickly and our industry kept on going. He added that the Semicon West 2021 has been rescheduled to Dec. 7-9, 2021, in San Francisco, USA.
Memory and storage in data economy
Indradeep Ghosh, Senior Director, Market Intelligence, Micron Technology, talked about memory and storage in the data economy. Electronics today has become a necessity of life. We have seen an acceleration in digital transformation. Memory and storage consumption has been accelerating. DRAM and NAND consumption has been growing per capita. DRAM and NAND revenue has been growing faster than the semiconductor industry.
Technology innovation has unlocked the data economy. Two major trends are AI and 5G. The data-centric cloud is moving to the intelligent edge, on to devices. They are accelerating innovation. New wave of innovation will transform the multiple industries over the next decade. These include mobility, healthcare, media and entertainment, agriculture, and industrial.
A few examples are connected smart vehicles and fully autonomous driving, remote health monitoring and early progonosis, remote operations and IoT in hospitals, immersive media, AR/VR and ubiquitous live streaming, AI-enabled user generated content, robotics, drones, satellite and soil sensors, end-to-end traceability for food safety and spoilage, and cloud control of machines, AR, video analytics, etc.
Data center DRAM is critical for compute-intensive apps. There will by ~13X AI server adoption by 2025. AI servers have ~6X the DRAM content of industry-standard servers. AI will become more pervasive. AI core use cases will be in business apps, content and collaborative, data management, gaming, media streaming, and web and app servers. AI use cases will also be in recommender systems, conversational technologies, image and video analytics, autonomous driving, cyber security, smart manufacturing, etc.
Data center NAND will be critical for data-hungry apps. NAND content on servers will more than double from 2020-2024. There will be 32 percent CAGR data center storage bit shipment growth. Few high-growth apps include structured data analytics, content apps, collaborative apps, app development and testing, etc. Performance will be across unstructured data analytics, media streaming, security, virtual desktop infrastructure (VDI), and engineering/technical apps.
Automotive will be the fastest growing memory and storage market. Content today is mostly infotainment driven. The ADAS adoption will be huge, with L1/L2 ~50 percent in 2020, and L3 <10 percent in 2025. Growing capabilities will be in large screen digital cockpit, ADAS L1/L2 and L3, event data recorder/driver monitoting, and telematic gateway.
In mobile, 5G will drive the smartphone content growth. There are growing 5G use cases. In photography and social, there are use cases such as 100+MP snapshots, 4K/8K video capture, triple picture/video capture, AI-enhanced real-time editing, 4K video livestream, LiDAR and advanced sensors. In entertainment and gaming, there are 4K display/immersive media, AR/VR shopping and gaming navigation, eSports/desktop-level gaming, etc. For healthcare and fitness, there are advanced sensors, AI-based health monitoring, etc.
For PCs, there will be new use cases driving a resurgence in demand. PCs saw double-digit growth in 2020. They are an essential device for WFH and remote learning. There are expanding use cases, such as video conferencing, apps to create, collaborate and productivity, entertainment, gaming and social.
The long-term DRAM bit demand CAGR will be of mid-high teens. The long-term NAND bit demand will be CAGR of approximately 30 percent.
State of EDA
Next, Jay Vleeschhouwer, MD, Griffin Securities, presented the state of EDA. The combined enterprise values of Cadence Design and Synopsys are ≈$80 billion, or more than 12x 2020 combined revenues, and almost 12x estimated 2021 revenues.
Five years ago, the combined enterprise values of Cadence, Synopsys and Mentor was ≈$16.1 billion. The material increase in value has been sustained by a combination of bookings growth, increasing backlog, increasing operating income (up 115 percent over the past half-decade), and increasing operating cash flow (up more than 115 percent over the past half-decade).
We estimate that EDA industry revenue increased by 11-12 percent in 2020 to more than $9.2 billion. The industry has continued to consolidate. Cadence and Synopsys – the Big 2 – accounted for ≈66 percent of industry revenues, as compared with ≈64 percent in 2015 and ≈53 percent in 2010. Mentor has also sustained its prior, pre-acquisition average share (19-20 percent), since it was acquired by Siemens in 2017. Mentor has shown good momentum in physical
verification (Calibre) and PCB. Synopsys-Cadence-Mentor-Ansys have nearly 90 percent of the industry revenues.
For 2021, we are estimating that Cadence’s revenues will increase 7 percent to $2.56 billion, and Synopsys’ EDA revenues will increase 7 percent to nearly $3.7 billion. Similarly, we are estimating that Ansys’ EDA business will increase by 7 percent to more than $360 million.
The earlier dip in EDA revenue was due to the recession. Japan has lost share, while Europe has gone sideways. Mentor has retained its revenue share. while Cadence and Synopsys have also increased their revenue.
The combined Big 2 EDA bookings were $5.89 billion in 2020, up ≈15 percent. We are estimating $5.93 billion for 2021 and more than $6.5 billion by 2023 – consistent with an expectation of better than mid-single-digit bookings growth and continued increases in backlog. The combined Big 2 backlog was $8.5 billion as of the end of 2020, up from $8 billion as of the end of 2019. We are estimating as much as $9 billion by the end of 2021. The combined EDA Big 2 operating income in 2020 was $2.075 billion, or 31.9 percent of revenues, vs. $1.574 billion in 2014, or 27.6 percent of revenues, and $967 million in 2015, or 24.4 percent of revenues.
For 2021, we are estimating combined income of $2.26 billion, or 32.5 percent of the estimated revenues, and $2.75 billion by 2023, or more than 35 percent of the estimated revenues.
Growth has been diverse, across many categories. This diverse base is expected to continue and be a driver. According to industry data, IC implementation, PCB, synthesis, analog/mixed-signal simulation, analysis, custom layout, and hardware-based verification have each had multiple consecutive periods of growth on a trailing-twelve-month (TTM) basis, plus improving trends for physical verification and RTL simulation.
The regular, co-inciding demand across multiple product categories by both the semiconductor and systems customers has been fundamentally conducive to EDA revenue growth – and, this phenomenon is very likely to continue. Each one of the EDA Big 4 – Synopsys, Cadence, Mentor and Ansys – participates in at least two of the growing categories. In physical verification, Mentor has dominated.
One of the most important product mix changes over the past 5-10 years has been growth of hardware-based verification (emulation and prototyping). The combined Cadence-Synopsys hardware revenues were more than $470 million in 2020, nearly doubling from 2015. Combined EDA IP revenues for Cadence and Synopsys were ≈$1.39 billion in 2020, (over 20 percent of combined revenue), vs. ≈ $1.1 billion in 2019, and more than ≈$620 million in 2015. The 2015-2020 CAGR for Big 2 core EDA software revenues (ex hardware and IP) was about ≈6 percent.
Two arms races
There are two arms races underway in technology: software development and silicon development. The investments in silicon development – by semiconductor companies, still the majority of EDA revenues, and the always important class of systems companies, e.g., Apple, Microsoft et al – are dependent upon EDA’s role as a source of essential technologies and services, and as such are sustaining the EDA industry’s revenue, income, and cash flow momentum.
The EDA industry growth has been sustained by growing demand among multiple EDA tool categories – as compared with earlier periods of more narrowly based growth. This has been, and is likely to remain, an important phenomenon, supported by the growth of semiconductor R&D budgets and systems customer product engineering budgets. These customer investments are in turn sustaining, and enabled by, EDA investments in R&D.
In 2020, the combined Cadence-Synopsys R&D was more than $2.35 billion (≈37 percent of revenues), vs. $2.116 billion in 2019, $1.429 billion in 2015 and $845 million in 2010. Cumulative combined R&D over the past decade (2010-2020) was more than $15.7 billion. We are estimating almost $2.5 billion for 2021 and more than $2.72 billion by 2023.
In semiconductor R&D, a composite of more than 25 semiconductor companies showed total R&D of $45.1 billion in 2019, up 2 percent. Intel accounted for ≈30 In semiconductor R&D, of this total. For the TTM ended 3Q20, total R&D was more than $46.8 billion, up almost 4 percent. The total R&D spending, excluding Intel, was up almost 4 percent in 2019 and ≈6 percent for the TTM ended 3Q20.
Among the semiconductor companies that have reported 2020 results, AMD’s R&D increased by 28 percent, Infineon’s by 29 percent, Intel’s by 1.5 percent, NXP’s by 5 percent, Nvidia’s by 39 percent (including Mellanox), Renesas’ by 2 percent, and ST’s by 3 percent. We have calculated that Intel’s commercial EDA spending accounts for as much as a high-single-digit percent of EDA industry revenues (more than $625 million in 2020). About three-fourths of its spending is with Synopsys, plus Cadence, Mentor and Ansys. There has been good bookings for the big 3 players over an 8-year period.
Part 2 continues later.
Semiconductor Industry Association (SIA) organized a conference on Memory Semiconductors: Market and Technology Trends. Memory semiconductors are a foundational and important sub-segment of the semiconductor market. There are various types of memory technologies, such as DRAM, NAND and NOR flash, and SRAM, etc.
According to the World Semiconductor Trade Statistics (WSTS), global sales of memory tied for the largest single share of any sub-product at 27 percent of the total market in 2020.
At the SIA conference, the speakers were Craig Stice, Chief Analyst at Omdia, Dr. Rainer Hoehler, VP and GM, Flash Solutions at Infineon Technologies, and Dr. Thomas Boone, VP for Defense and Aerospace at Spin Memory. The session was moderated by Falan Yinug, Director of Industry Statistics and Economy Policy at SIA.
Craig Stice, Omdia, said DRAM and NAND are the high-volume, commodity memory semi-components, working together in a system (such as PC, server, smartphone), but for different reasons. DRAM manages your data, requires power (volatile). NAND stores your data, does not require power (non-volatile). Memory is in just about all electronics.
Enterprise (server and data center), PC, and mobile make up 80+ percent of the DRAM and NAND demand. TVs, game consoles, removeable storage cards, automotive, smart speakers, other CE, industrial and medical applications, military, etc., also use memory.
Memory in 2020
In 2020, DRAM & NAND combined revenue reached ~$120 billion. In 2020, the total semiconductor market reached revenue of ~$470 billion. DRAM and NAND made up ~25 percent of worldwide total semiconductor revenue. DRAM and NAND are a significant influence on the overall semiconductor market.
The memory industry is well understood to be cyclical, driven by supply and demand. Most recently, following record revenues in both 2017 and 2018, 2019 was a historical down cycle for both DRAM and NAND as demand grew stagnant, memory bit growth persisted, and inventory levels increased resulting in aggressive average selling price (ASP) declines.
In the memory industry, what goes up, typically comes down, then back up. In 2020, the memory industry bounced back, even during a worldwide pandemic. It takes a lot of infrastructure to support work-from-home, educate-from-home, and entertain-from-home environments. Looking ahead, we expect cycles to continue as strong revenue years is re-invested into capacity expansion.
Reviewing 2020, he said that the year 2020 had been projected to be a rebound year from a disastrous 2019. Then Covid-19 hit! Scares of memory production shut-downs, logistical issues, and demand going stagnant put the market on high caution. Smartphones took the early big hit in demand. ~200 million smartphones were taken out of the total 2020 forecast in early 1Q20. However, memory demand from datacenters erupted due to ‘at-home’ environments.
PC markets flourished as corporate, consumers, and educational markets were upgrading PCs for at-home use. At-home entertainment spiked (TV, gaming). Memory pricing increased in 1H20. By 2H20, the mobile markets had bounced back, PC demand continued to outperform, but the enterprise markets went into an inventory management phase and demand slowed. Memory pricing began to decline in 2H20.
In the short-term (2021), consumer will see strong home entertainment demand, such as TV, set-top-box, smart speaker, etc. In graphics, there will be strong game console demand (new SSD demand), and GPU demand will be boosted by cryptocurrency mining. In mobile, there will be recovery of smartphone, and continued low/mid-tier priced 5G smartphone launches. In server, there will be recovery of enterprise server, and Intel Ice Lake launch. However, cloud server demand is slower than last year. In PC, there will be strong demand for educational and gaming notebooks. There will be caution for a possible softening in 2H21.
In the mid-term (2022-2024), in consumer, there will be mild correction in consumer device, and continued IoT demand. In gaming, there will be mild correction in game consoles and GPUs. In mobile, there will be saturated 5G smartphone set, and growth in low-end segment from emerging countries. In server, there will be DDR5 replacement demand, and SSD buildout will continue. There will be edge server and AI momentum, as well. In PC, there will be some stability from notebooks.
In the long-term (beyond 2024), in consumer, there will be continued demand of consumer electronic devices. In gaming, there will be multi-year cycle of new game consoles and GPUs. In mobile, there will be replacement demand of smartphones. In server, edge computing and AI momentum will continue. There will be HDD cold storage replacement with SSD. In PC, there will be replacement demand of the enterprise segment, and continued positive momentum of the low-end segment.
DRAM technology and manufacturing
In DRAM technology, DRAM has historically gone through ‘die shrinks’. The printed x/y memory cell pattern on the silicon wafer gets smaller and smaller with next- generation lithography nodes and architectural changes. Next-generation die will bring higher capacity, faster speeds, lower power, etc. There will be more die per wafer, more bits per wafer = better cost per wafer. However, as DRAM moves down the roadmap, next generations are becoming more challenging (longer time between generations), more expensive, and not as big of a cost savings.
In DRAM manufacturing, three manufactures make up roughly 95 percent of all DRAM revenue, namely, Samsung, SK Hynix, and Micron. By comparison, in 2008 the same three made up only ~60 percent of the revenue share. Very volatile markets lead to strong supply growth, fierce competition, and ultimately M&A.
With only three major players, and with DRAM technology getting challenging and longer to get to next generations, DRAM output is much more stable. CXMT will be the new DRAM manufacture starting up in China.
NAND technology and manufacturing
For NAND technology, beginning in 2017, the NAND industry went through a 2D to 3D NAND architecture change. NAND performed die shrinks as far as it could go, so went vertical. Think of it as a single-story home to a high-rise apartment. There was same footprint, but more capacity. These vertical layers create more bits per wafer, better cost, higher capacity (with more layers). Today, the NAND industry is on 96-layer 3D NAND, with 128-layer in early production. There is no declaration of how high the industry can stack, but it will become more challenging to achieve ideal yields.
As for NAND manufacturing, Samsung, SK Hynix, and Micron also manufacture NAND. The three comprise of roughly 67 percent of global NAND revenue. Kioxia (formerly Toshiba Memory) and Western Digital have a partnership, near splitting wafer output. Intel is being acquired by SK Hynix beginning late 2021. YMTC is the new company out of Wuhan, China, but it still has very small output.
DRAM and NAND combined is a $100+ billion industry, but very cyclical and driven by supply and demand. DRAM and NAND make up ~25 percent of the total semiconductor industry. PCs, mobile, and enterprise (data center/server) make up roughly 80 percent of all DRAM and NAND demand.
Overall, the memory industry is expected to continue to grow to meet demand from consumer electronics, hyperscale data centers, AI, IoT, autonomous vehicles. However, technology challenges will only persist in the years ahead. DRAM die shrinks are becoming more challenging. With three manufacturers now making up most of the DRAM output, supply can be more stable and predictable.
NAND has gone vertical, now at 96-layers and moving to 128-layers. DRAM and NAND are a significant influence on the total semiconductor market, either up or down.
Infineon offers unique portfolio
Dr. Rainer Hoehler, Infineon, stated they offer a unique portfolio that links the real and the digital worlds. Infineon offers coin-cell-powered devices, battery-powered devices, memory for power supplies, industrial IoT, drives, smart homes, consumer IoT devices, 5G, and automotives.
Applications rely on safe, secure and reliable memory solutions. There is reliable boot code, secure storage, over-the-air update. The solutions offer instant-on performance, functional safety and security. Compact and low power, features include longevity (15+yrs) and reliability at high temperatures (>125˚C). Infineon serves the automotive, communications, industrial, and wearables segments.
Innovations in memory solutions are enabling emerging applications. These are across automotive, industrial and medical/A&D, and communications domains. Emerging applications require differentiated solutions. The Infineon SEMPER NOR Flash features integrated functional safety and security, and integrated compute core. The serial/parallel NOR Flash features automotive-grade reliability, and high random access performance at low power. nvSRAM combines SRAM performance with power-loss data protection.
The EXCELON F-RAM has NoDelay writes, and 100 trillion read/write cycles. SRAM offers best standby power and lowest SER with embedded ECC. Also, HyperRAM offers low pin-count, and common HyperBus/Octal interface with NOR-Flash. Infineon also offers the Semper Flash architecture, or, the Flash you can trust.
Electronics are pervasive in the real world. Our lives increasingly depend on them. Apps are diversifying, and different technologies and solutions are required. Dependability is the key. Functional safety and security against cyberattacks is required.
Focus on STT-mRAM
Spin Memory is a US company located in Fremont, CA developing perpendicular spin transfer torque magnetic random access memory (pSTT-MRAM). It is developing technology to provide embedded NVM, SRAM, and ultimately, DRAM.
Spin Memory owns a class 100 cleanroom ‘back-end’ manufacturing facility. It is currently developing radiation hardened memory solutions to support US military and space applications. STT-MRAM is inherently rad-hard!
Explaining STT-MRAM, Dr. Thomas Boone said MRAM is magneto-resistive RAM. STT is spin transfer torque. Electron spin sets free layer polarization. ST-MRAM using pMTJ is latest MRAM generation. Bitcell uses 1 transistor + 1 MTJ, a very dense configuration. There are attributes such as non-volatile, high-speed read and write, high endurance, and easy integration in BEOL, with no impact on FEOL. There is perpendicular magnetic tunnel junction (pMTJ), as well.
The US Government and DoD requires advanced Rad-Hard microelectronics memory for strategic and space applications. STT-MRAM is intrinsically radiation immune. It is useful for space probe and satellite apps. Spin Memory is targeting trusted and assured foundries. Spin Memory is enabling STT-MRAM IP and resources, including CONUS 200mm factory in Fremont, CA. Space-based systems also need a new memory. Spin’s patented IP engineered MRAM is set to challenge DRAM/SRAM.
At the IFS 2021 annual semiconductor industry update and forecast today, in London, UK, Malcolm Penn, Founder and CEO, Future Horizons, presented the findings for the global semiconductor industry for 2021.
There was hope in 2020, with enforced stay-at-home and WFH, etc. Demand for IT products, telecoms, etc., increased. The GDP slump was painful, although it was not a classic meltdown. Strong rebound is likely once the weather turns milder and more get vaccinated against Covid-19. The world GDP was -4.4 percent now, vs. -4.9 percent in June 2020.
Strong rebound is expected in 2021. Supply in the semiconductor market is systemic and strategic. GDP had collapsed in 2020, but end demand skyrocketed offsetting the impact. The global IC unit sales by month is remarkably consistent at +8 percent per year for 2021. In 2020, there was +5.6 percent growth, despite 5.1 percent GDP decline. We bounced back very well. Also, units/silicon area was seeing ~2 percent growth. Wafer vs. time PMCC is 0.9382, IC units vs. time PMCC is 0.09698, and IC units vs. wafers PMCC was 0.9491. PMCC shows how matched the variables are.
Leading edge fab capacity is very carefully managed. New capacity, and demand, are not forecast driven. Semiconductor device and total capex market dynamics see cautious capex spend accelerating, but not near overheating, at 15.6 percent in 2020. TSMC is looking at $28 billion capex spend, and Samsung is looking at $30 billion capex spend in 2021. Dollars per square inch has also remained constant.
The IC ASP cycle is showing fast decline and slow recovery. We should be looking at another oscillating cycle. The economy remains hostage to Covid-19. IC units growth spurt is in progress, holding up strong strain. Fab capacity has been sold out, with no near-term relief in prospect. ASPs recovery could happen in 2021. The economic outlook is fragile, but all the other fundamentals are strong. We need to spend more, and see the results by January 2022. There is likely memory price recovery in H2-2021.
The outlook was +10.2 percent for 2021. Double digit growth is inevitable, with 10 percent as minimum. In March 2020, there was potential for strong H2-2020 rebound. We presented the outlook for global semiconductor industry was +1.4 percent. Besides automotives and smartphones, the industry hardly blinked. Q4 sent everyone scrambling for capacity. Q4-2020 was the start of the next chip industry super cycle. Q4-2020 was a one trillion unit quarter.
The improving industry momentum will be going into 2021. There will be more stable economic footing following the US election, the UK-EU trade deal, and less hostile US-China trade embargos and tariff wars. Sound industry fundamentals are providing stable growth platform. Three concers are living with China in an increasingly decoupled world, there is uncertain economic roadmap to unwinding fiscal support, and delays in returning to post-Covid-19 normality. The balance of semiconductors growth has more upsides than downs.
The current outlook for the global semiconductor industry is +18 percent growth for 2021. There is no tight capacity relief before 2022 at the earliest. DRAMs are moving ahead. Scaling is now a continuous improvement vs. full node progress. Samsung is now using EUV for 10nm class D1x process. Its the same with 3D NAND. Micron is ahead with 176 layers. There is a 4D NAND structure in place now, especially from Micron. There is talk about 256 layers down the road.
Emerging memory highlights include XPoint technology and products, ReRAM from Adesto, CBRAM, Dialog, etc. TSMC roadmap is really impressive, in logic. N5 is already in full node, for Apple. N5+ is likely in Q1-2021. In advanced R&D, manufacturing and market, there are new transistor structures and materials. Scaling continues through EUV breakthrough, at TSMC, with Apple having the power of a sugar daddy.
EV as growth driver
Key industry growth drivers include AI and neural networks, electric vehicles, etc. Apple and Foxconn will change the automotive world, as EV is the means to the end. It also opens the door to sub-contracting and economy of scale. Tesla was first one off the block, and had a massive lead. There are 260mm2 inhouse designed SoCs.
Foxconn wants to be the Android of automobiles. It is looking at 10 percent global market share for EV platforms by 2025. First Foxconn open platform EVs are likely by 2022. It has a JV with Taiwan’s Yulon (after Hotai Motors) to develop EV cars. It is bidding to buy Silterra.
There will be a cleaner hydrogen alternative, with no charging points, faster refuelling, and build off the existing LPG architecture. Asia will be market driver with Toyota Mirai, Hyundai Nexo and Honda Clarity.
Automobiles is now just like any other system. It is easier for Apple and Google to enter the autonomous car market, with production outsourced to TSMC and Foxconn. The US car industry is now lobbying Joe Biden to get its chips. There is also disruptive innovation in the OEM direct model. Silicon is the founding block for technology. The ecosystem is well developed now.
Key takeaways for 2021 are that EVs are causing massive automobile industry disruption. There is huge pent-up end-user demand, from cars to holidays, and everything in between. Remote working, videoconferencing, voice activation, etc., are here to stay. There is massive acceleration of personal health and medical, dwarfing everything. There is no shortage of technology, and Moore’s Law is not dead, or sick. The rebound will be strong. The US-China friction has to be dealt with. While China races ahead, the US near-term outlook remains fraught. China GDP grew 1.9 percent in 2020. China economy may overtake the USA by 2028.
SEMI, USA organized a webinar today on market data resources and equipment materials outlook for 2021. Ed Hall, Senior Manager, Information Products (Market Reports/Standards), SEMI, presented on the SEMI market data resources. A demonstration of the global semiconductor packaging materials outlook (GSPMO) report and the silicon wafer monitor report was shown.
China, Asia lead recovery
Clark Tseng, Director, Industry Research & Statistics, SEMI, presented the semiconductor equipment/materials market outlook. He said that there had been an improving 2020 for the world economy, and hopes for a promising 2021. The global manufacturing PMI rebound continued in September 2021. China and Asia are leading the economic recovery. China exports continue to expand in Q3-20. The China PMI has been above 50 in the last seven months. There has also been strong rebound of the automotive sector.
The forecast for the global semiconductor market was revised from 8 percent to 3.5 percent for 2020. For 2021, the semiconductor forecast average is +9 percent. That’s an encouraging sign for recovery! Covid-19 accelerated digital transformation of many industries and services, globally. Semiconductors have been resilient to the pandemic, and benefits from the Covid-19-driven transformations.
Restriction on Huawei will have short-term impact on the supply chain. This has accelerated the pace of localization for major Chinese OEMs. The rise of techno-nationalism will accelerate the reshoring and decoupling of the supply chain, including the upstream semiconductor manufacturing. North American billings strength has continued to Q3-20. Sept. 2020 billings set new record of $2.75 billion.
WFE market expands
Wafer fab equipment market size has expanded from ~$30 billion in H1-2010 to over ~$50 billion in the past three years. WFE forecast by product segment shows the overall market is expected to show healthy growth for next few years. DRAM and NAND spending are likely to increase mid-teens percent in 2021, and remain robust for the next few years. Foundry and logic spending is to reach ~$30 billion in 2022 and 2023.
Test equipment market is likely to grow over10 percent in 2020, and further in 2021. Growth is driven by 5G and HPC. Assembly and packaging equipment are also forecasted to grow over 10 percent in 2020, and 8 percent in 2021. This is driven by advanced packaging, etc. SEMI revised the 2020 forecast from $61 billion to $65 billion given strong H1-20 equipment billings. Foundry and logic remain strong in H2-20, while memory recovery may be slower than expected. China spending has been strong in H-20 and continues to H2-20.
Silicon wafer market moves upward
Regarding silicon wafer shipment trend, YoY through Q3-20 wafer shipment increased 2.7 percent. Q2-20 shipment rebounded 8 percent QoQ. Materials market exceeded $50 billion since 2018, driven by shipment growth and advanced process requirements. SEMI expects 6 percent growth in 2021.
Wafer fab materials are projected to decline 0.5 percent in 2020, but rebound 7 percent in 2021. Packaging materials will grow 2 percent in 2020, and 5 percent in 2021. Substrates continue to show strength. Leadframe is forecast to drop 3 percent in 2020, but rebound 5 percent in 2021. Ceramic packages and encapsulation resins will see soft 2020, but recover in 2021.
SK Hynix thrives, SMIC fumbles
There have developments regarding restrictions to SMIC and SK Hynix’s acquisition of Intel NAND flash. SMIC accounts for 7.5 percent of global foundry capacity and higher percent in mature nodes. SMIC accounted for 7.5 percent of WFE market in 2020. SMIC recently announced the N+1 tape out. SEMI expects significant decline in SMIC’s fab spending in 2021. Opportunities have also emerged for non-US tool suppliers as SMIC seeks to diversify its suppliers.
SK Hynix acquisition of Intel NAND flash has had no immediate impact on the supply and demand. The acquisition provides SK Hynix access to the enterprise SSD market. Dalian fab provides instant capacity addition (over 80K), economy of scale, and room for future expansion. This is a big plus for the Korean equipment suppliers in the long run, for future capacity expansion and technology migration. Tsinghua Unigroup was considered the forerunner of Intel’s NAND flash business.
The foundry market is expected to see 20 percent growth in 2020, driven by digital transformation. All technology nodes are seeing strong growth, except 28nm. As SMIC customers are exploring alternative routes, SEMI has seen sudden foundry demand spike in recent months. 200mm capacity is in tight supply with demand from PMIC, DDI, MCU, sensors, etc.
DRAM to recover
DRAM price decline will continue into Q4-20, with 10 percent growth due to elevated inventory. There is a pause of server-related (memory) procurement in H2-20, cloud/server demand is poised to improve in the coming quarters from USA and China customers. Mobile DRAM demand is a bit better right now. Higher 5G phone penetration will provide some upside in 2021. PC DRAM is better than expected, as laptop/chromebook strength will continue into Q1-21. DRAM price is expected to bottom out and recover in Q1-21.
Overall, SEMI remains positive on 2021 forecasts for electronics and IC sales. Data centers, HPC and AI, are still the key industry drivers. Laptop/server demand is likely to continue into 2021. 5G adoption may be slower than expected. Massive government stimulus packages will continue to boost the recovery. Semiconductor equipment and material markets are expected to grow in 2021.
The electronic design automation (EDA) industry revenue increased 12.6 percent in Q2-2020 to $2,783.9 million, compared to $2,472.1 million in Q2-2019, with most categories logging double-digit increases, as per the Electronic System Design (ESD) Alliance Market Statistics Service (MSS).
The four-quarter moving average, which compares the most recent four quarters to the prior four quarters, increased by 6.7 percent. The ESD Alliance is a SEMI Technology Community.
Dr. Walden (Wally) C. Rhines, Executive Sponsor, SEMI EDA Market Statistics Service, President and CEO, Cornami, and CEO Emeritus, Mentor, A Siemens Business, said that the EDA industry is experiencing amazingly strong growth right now, at least through the second quarter.
“We just reported 12.6 percent worldwide growth in revenue compared to the same quarter as last year. The last four quarters show growth of 6.7 percent. In the second quarter of 2020, every category tracked by the ESDA Market Statistics Program grew in double digits, except PCB design and services. Even so, PCB is still on track to be the fastest growing segment in 2021 with 12.4 percent growth in the last 12 months.”
Right now, all the segments are increasing. Will this trend continue, going forward?
Dr. Rhines said: “Of course, no one can predict the future. But, the underlying fundamentals causing current growth suggest that this is not a short-term effect. The biggest contributor right now is the adoption of EDA software by companies that have not historically designed their own electronics. That includes the IT community of companies like Google, Facebook, Amazon, Alibaba, and many more. In addition, the other systems companies in areas like automotive electronics are doing their own chip and board designs, while continuing their dependence upon tier one providers like Bosch, Denso, and many more.”
So, what is the growth likely for EDA during 2021? According to him, Japan grew at a very strong 9 percent rate in Q2-20 versus Q2-19. But, for the last 12 months, it has been flat. Korea EDA revenue decreased about 10 percent over the past 12 months, compared to prior years, and was flat in Q2-20 vs. Q2-19.
And, how are the semiconductor markets in Korea and Japan looking right now? Dr. Rhines noted that Japan is relatively flat. Korea has easier comparisons with last year, since the decrease in semiconductor revenue in 2019 was heavily influenced by memory price declines. He would expect that Korea will grow its semiconductor sales more than the overall world average in 2020 due to some recovery in memory pricing, influenced by the strong demand for server capacity in data centers.
Next, will logic performance improvement at fixed power slow down in 2021? How do you get around that? Dr. Rhines said that semiconductor logic revenue was relatively flat in 2019 despite the overall semiconductor market decline. It appears to be continuing that trend in 2020. The year 2021 will depend upon a post Covid-19 economic recovery. That is by no means certain, and it’s probably influenced by other factors, such as the elections in the USA.
Will there also be more heterogeneous integration, enabled by 3D technologies? “Absolutely! Heterogeneous integration is growing rapidly,” noted Dr. Rhines. And, multi-chip 2D and 3D packaging has made many new capabilities possible. Integration of PCB layout tools with IC design environments also helped. The support of foundries, like TSMC, with their “3D Fabric”, has helped too. Chiplets are an interesting extension to this packaging capability. It is interesting to see what AMD and Intel are doing in this space.
NVM and edge AI on the rise
Further, does the industry see the emerging non-volatile memories on the rise? He added that non-volatile memories are currently leading the transistor cost learning curve for the semiconductor industry. About 512 layers in flash memories is achievable, making for amazing NAND flash capacity in a single package. At this same time, it is a period of growing interest in new memory process technologies, like MRAM, ReRAM, FRAM, and more. Cost per bit continues on the long-term learning curve, and the continued doubling of total memory storage, both appear very predictable.
On the same token, I asked for his thoughts on the edge AI chip industry, going forward. As per Dr. Rhines, edge AI is entering a new wave of growth. It is inevitable that the intelligence in the cloud will make its way downward to embedded systems. It always has in the past as silicon capability allows us to compute locally that which we used to compute centrally in mainframes or servers.
“Dozens of new post-Von Neumann neuromorphic computing architectures have been funded as chip startups starting in 2017 and the pace continues at about $2 billion of venture investment per year in these companies. Working for one of these companies, Cornami, that promises orders of magnitude further performance and power dissipation improvements gives me some visibility into this trend. It is not slowing down.”
Finally, does the NAND industry need to consolidate to generate sufficient returns? He said that most semiconductor mergers and acquisitions are no longer driven by manufacturing economies of scale, unlike the 1970s and 1980s. Memory manufacturing efficiency does, however, depend upon scale. Samsung has nearly 30 percent of the market, and is very profitable, although, the commodity nature of memory makes the revenue and pricing more volatile than non-memory semiconductor products.
“Behind Samsung, we have Kioxia (formerly Toshiba), Micron, Western Digital (SanDisk) and SK Hynix. I suspect that harvesting economies of scale from merging any of these companies would be difficult because of the differences in products, processes and geographic locations. But, it could certainly happen!”
TrendForce has provided its forecast of 10 key trends in the tech industry for 2021.
As the DRAM industry officially enters the EUV era, NAND Flash stacking technology advances past 150L
The three major DRAM suppliers, Samsung, SK Hynix, and Micron, will not only continue their transition towards the 1Znm and 1alpha nm process technologies, but also formally introduce the EUV era, with Samsung leading the charge, in 2021. DRAM suppliers will gradually replace their existing double patterning technologies in order to optimize their cost structure and manufacturing efficiency.
After NAND Flash suppliers managed to push memory stacking technology past 100 layers in 2020, they will be aiming for 150 layers and above in 2021 and improving single-die capacity from 256/512Gb to 512Gb/1Tb. Consumers will be able to adopt higher-density NAND Flash products through the suppliers’ efforts to optimize chip costs.
While PCIe Gen 3 is currently the dominant bus interface for SSDs, PCIe Gen 4 will start gaining increased market share in 2021 owing to its integration in PS5, Xbox Series X/S, and motherboards featuring Intel’s new microarchitecture. The new interface is indispensable for fulfilling the massive data transfer demand from high-end PCs, servers, and HPC data centers.
Mobile network operators will step up their 5G base station build-out while Japan/Korea look ahead to 6G
The 5G Implementation Guidelines: SA Option 2, released by the GSMA in June 2020, delves into great technical details regarding 5G deployment, both for mobile network operators and from a global perspective. Operators are expected to implement 5G standalone architectures (SA) on a large scale in 2021.
In addition to delivering connections with high speed and high bandwidth, 5G SA architectures will allow operators to customize their networks according to user applications and adapt to workloads that require ultra-low latency. However, even as 5G rollout is underway, Japan-based NTT DoCoMo and Korea-based SK Telecom are already focusing on 6G deployment, since 6G allows for various emerging applications in XR (including VR, AR, MR, and 8K and above resolutions), lifelike holographic communications, WFH, remote access, telemedicine, and distance education.
Internet of Things evolves into Intelligence of Things as AI-enabled devices move closer to autonomy
In 2021, deep AI integration will be the primary value added to IoT, whose definition will evolve from Internet of Things to Intelligence of Things. Innovations in tools such as deep learning and computer vision will bring about a total upgrade for IoT software and hardware applications. Taking into account industry dynamics, economic stimulus, and remote access demand, IoT is expected to see large-scale adoption across certain major verticals, namely, smart manufacturing and smart healthcare.
With regards to smart manufacturing, the introduction of contactless technology is expected to speed up the arrival of industry 4.0. As smart factories pursue resilience, flexibility, and efficiency, AI integration will equip edge devices, such as cobots and drones, with even more precision and inspection capabilities, thereby transforming automation into autonomy. On the smart healthcare front, AI adoption can transform existing medical datasets into enablers of process optimization and service area extension.
For instance, AI integration delivers faster thermal image recognition that can support the clinical decision-making process, telemedicine, and surgical assistance applications. These aforementioned applications are expected to serve as crucial functions fulfilled by AI-enabled medical IoT in diverse settings ranging from smart clinics to telemedicine centers.
Integration between AR glasses and smartphones will kick-start a wave of cross-platform applications
AR glasses will move towards a smartphone-connected design in 2021 in which the smartphone serves as the computing platform for the glasses. This design allows for significant reduction in cost and weight for AR glasses. In particular, as the 5G network environment becomes more mature in 2021, the integration of 5G smartphones and AR glasses will enable the latter to not only run AR apps more smoothly, but also fulfill advanced personal audio-visual entertainment functionalities through leveraging the added computing power of smartphones. As a result, smartphone brands and mobile network operators are expected to venture into the AR glasses market on a large scale in 2021.
A crucial part of autonomous driving, driver monitoring systems (DMS) will skyrocket in popularity
Automotive safety technology has evolved from an application for car exteriors to one for car interiors, while sensing technology is moving towards a future where it integrates driver status monitoring with external environmental readings. Similarly, automotive AI integration is evolving past its existing entertainment and user assistance functions, into an indispensable enabler of automotive safety.
In light of the string of traffic accidents in which the drivers ignored road conditions due to their overreliance on ADAS (advanced driver assistance systems), which have recently skyrocketed in adoption rate, the market is once again paying close attention to driver monitoring functions.
In the future, the main thrust of driver monitoring functions will be focused on the development of more active, reliable, and accurate camera systems. By detecting the driver’s drowsiness and attention through iris tracking and behavioral monitoring, these systems are able to identify in real time whether the driver is tired, distracted, or driving improperly.
As such, DMS (driver monitoring systems) have become an absolute necessity in the development of ADS (autonomous driving systems), since DMS must serve multiple functions simultaneously, including real-time detection/notification, driver capability assessment, and takeover of driving controls whenever necessary. Vehicles with DMS integration are expected to enter mass production in the near future.
Foldable displays will see adoption in more devices as a means of upping screen real estate
As foldable phones progressed from concept to product in 2019, certain smartphone brands successively released their own foldable phones to test the waters. Although these phones’ sell-through performances have so far been mediocre owing to their relatively high costs – and, by extension, retail prices – they are still able to generate much buzz in the mature and saturated smartphone market. In the next few years, as panel makers gradually expand their flexible AMOLED production capacities, smartphone brands will continue to focus on their development of foldable phones.
Furthermore, foldable functionality has been seeing increasing penetration in other devices as well, specifically notebook computers. With Intel and Microsoft leading the charge, various manufacturers have each released their own dual-display notebook offerings. In the same vein, foldable products with single flexible AMOLED displays are set to become the next hot topic.
Notebooks with foldable displays will likely enter the market in 2021. As an innovative flexible display application and as a product category that features flexible displays much larger than previous applications, the integration of foldable displays in notebooks is expected to expend manufacturers’ flexible AMOLED production capacity to some degree.
Mini LED and QD-OLED will become viable alternatives to white OLED
Competition between the display technologies is expected to heat up in the high-end TV market in 2021. In particular, Mini LED backlighting enables LCD TVs to have finer control over their backlight zones and therefore deeper display contrast compared with current mainstream TVs. Spearheaded by market leader Samsung, LCD TVs with Mini LED backlighting are competitive with their white OLED counterparts while offering similar specs and performances.
Furthermore, given their superior cost-effectiveness, Mini LED is expected to emerge as a strong alternative to white OLED as a display technology. On the other hand, Samsung Display (SDC) is betting on its new QD OLED technology as a point of technological differentiation from its competitors, as SDC is ending its LCD manufacturing operations. SDC will look to set the new gold standard in TV specs with its QD OLED technology, which is superior to white OLED in terms of color saturation. TrendForce expects the high-end TV market to exhibit a cutthroat new competitive landscape in 2H21.
Advanced packaging will go full steam ahead in HPC and AiP
The development of advanced packaging technology has not slowed down this year despite the impact of the Covid-19 pandemic. As the various manufacturers release HPC chips and AiP (antenna in package) modules, semiconductor companies such as TSMC, Intel, ASE, and Amkor are eager to participate in the burgeoning advanced packaging industry as well. With regards to HPC chip packaging, due to these chips’ increased demand on I/O lead density, the demand on interposers, which are used in chip packaging, has increased correspondingly as well.
TSMC and Intel have each released their new chip packaging architectures, branded 3D fabric and Hybrid Bonding, respectively, while gradually evolving their third-generation packaging technologies (CoWoS for TSMC and EMIB for Intel), to fourth-generation CoWoS and Co-EMIB technologies.
In 2021, the two foundries will be looking to benefit from high-end 2.5D and 3D chip packaging demand. With regards to AiP module packaging, after Qualcomm released its first QTM products in 2018, MediaTek and Apple subsequently collaborated with related OSAT companies, including ASE and Amkor. Through these collaborations, MediaTek and Apple hoped to make headways in the R&D of mainstream flip chip packaging, which is a relatively low-cost technology.
AiP is expected to see gradual integration in 5G mmWave devices starting in 2021. Driven by 5G communications and network connectivity demand, AiP modules are expected to first reach the smartphone market and subsequently the automotive and tablet markets.
Chipmakers will pursue shares in the AIoT market through an accelerated expansionary strategy
With the rapid development of IoT, 5G, AI, and cloud/edge computing, chipmakers’ strategies have evolved from singular products, to product lineups, and finally to product solutions, thereby creating a comprehensive and granular chip ecosystem. Looking at the development of major chipmakers in recent years from a broad perspective, the continuous vertical integration of these companies have resulted in an oligopolistic industry, in which localized competition is more intense than ever.
Furthermore, as 5G commercialization generates diverse application demands for various use cases, chipmakers are now offering full service vertical solutions, ranging from chip design to software/hardware platform integration, in response to the vast commercial opportunities brought about by the rapid development of the AIoT industry. On the other hand, chipmakers who were unable to position themselves in time according to market needs will likely find themselves exposed to the risk of overreliance on a single market.
Active matrix Micro LED TVs will make their highly anticipated debut in the consumer electronics market
The release of large-sized micro LED displays by Samsung, LG, Sony, and Lumens in recent years marked the start of micro LED integration in large-sized display development. As micro LED application in large-sized displays gradually matures, Samsung is expected to be the first in the industry to release its active matrix micro LED TVs, therefore cementing year 2021 as the first year of micro LED integration in TVs.
Active matrix addresses pixels by making use of the display’s TFT glass backplane, and since the IC design of active matrix is relatively simple, this addressing scheme therefore requires a relative low amount of routing. In particular, active matrix driver ICs require PWM functionality and MOSFET switches in order to stabilize the electrical current driving micro LED displays, necessitating a new and extremely expensive R&D process for such ICs. Therefore, for micro LED manufacturers, their greatest challenges at the moment in pushing micro LED to the end devices market lie in technology and cost.
Please note: These are all the predictions from TrendForce, Taiwan for 2021.
Yole Developpement, France, organized a session on the NAND landscape. It also looked at how Kioxia, and the other leading memory players are doing.
Walt Coon, VP, NAND and Memory Research, Yole Developpement, provided the NAND market overview. The market is dominated by six primary suppliers — Samsung, Kioxia, Western Digital, SK Hynix, Micron and Intel. Samsung leads the market with 33 percent share. All the major players contribute 99 percent of the market.
Yangtze Memory Technologies Co. Ltd or YMTC has now entered the market as the seventh NAND supplier. They are running at 20,000 wafers per month. The company has Fab 1 ramping in Wuhan, supporting up to ~100K wafers per month. Wuhan Fab 2 should ramp in 2022. A third fab is possible, either in Wuhan or Chengdu, ramping in 2023.
The NAND market has experienced tremendous growth over the past two decades. Fundamental demand for NAND is strong, and continues to expand as the technology-driven price declines enable new markets. 2020 was poised to be great year, before Covid-19 struck. There is working and learning from home. Revenue should be down in Q3 and Q4 this year. The industry profitability has remained limited. Intense competition has led to price volatility and periods of losses. Covid-19 will likely push down costs.
Yole had expected revenue to reach $60 billion in 2020, and shipments to grow by 30 percent, before Covid-19 struck. The current outlook sees revenue to reach $55 billion, and shipments to grow 27 percent. In the longer term, we expect the NAND market to expand. NAND will continue to grow into traditional HDD sockets as NAND costs decline. Data generation and storage trends are showing no signs of decline.
Now, let’s look at Kioxia’s performance in scale and market share. Toshiba Memory Corp. was renamed to Kioxia in 2019. Kioxia holds 50.1 percent of Flash Ventures, a JV with Western Digital that develops and manufactures NAND memory for both companies. There are manufacturing locations at Kitakami and Yokkaichi, in Japan.
NAND wafer capacity is the highest in the industry. Samsung is the leader, and YMTC is poised to grow in 2020-21. The JV NAND capacity between Kioxia and WD is today the highest in the industry. Kioxia’s NAND revenue generally moved inline with the industry. In the context of the total memory market (including DRAM, NOR, 3D, XPoint), Kioxia’s position is much smaller, at no. 4, with less than 10 percent share.
Looking at the 3D NAND roadmap, Samsung has been the first to get off. Kioxia/WD’s first three 3D NAND generations have been competitive with the other suppliers in terms of performance, memory, density, die size, yields, cost per bit, etc. The JV is expected to introduce the CMOS under array or CuA architecture, and shrink lithography on 160-layer (vs. standard 176-layer), to increase the GB/wager. It will improve its relative competitive position. The 160-layer is presumably to increase the tool re-use.
Kioxia/WD’s JV capex has increased in the recent years to support the 3D NAND ramp up and new wafer capacity to offset the reduced wafer throughput. The JV’s capital intensity has consistently been below the industry (and best-in-class), evidence of the strong execution and highlighting the importance of scale.
Samsung is the undisputed leader in SSD and mobile segments (largest markets), followed by Kioxia. Intel participates in the SSD segments exclusively, with strength in the enterprise/data center segments. Kioxia (formerly, Toshiba Memory), has consistently been profitable on an annual basis, with the downturn in 2019 as the only exception. The market (and Kioxia) is still subject to volatility from industry investment cycles.
The NAND market has grown dramatically over the last 20 years. Fundamental demand for NAND is strong. TAM is likely to expand as HDD sockets continue the transition to NAND. The near-term market headwinds exist from Covid-19. The longer-term outlook is complicated by the emergence of YMTC. As of now, six major players exist today, with YMTC on the horizon.
The NAND industry needs to consolidate to generate sufficient returns, following the DRAM industry (6 major suppliers down to 3), and now, currently much more profitable, relative to NAND.
Kioxia is a leading player in the NAND market, with the second-highest market share. It has had consistent presence in the industry, since inventing the technology in the late 1980s. Kioxia participates in a mutually-beneficial partnership with Western Digital. This provides the benefits of scale and shared technology investments.
Due to a lack of DRAM, Kioxia is less diversifed than its competitors — Samsung, SK Hynix and Micron. It has a lower overall memory market share of less than 10 percent. Kioxia’s profitability has improved over the past 10 years. The company is still subject to the volatility of the memory market.