2024年全球半导体产业展望（英）.pdf

2024 global semiconductor industry outlook 2Executive summary 3Generative AI and the next wave of chip supply 4Making smart manufacturing smarter 6Semiconductor assembly and test venture into new geographies 8Bolstering cybersecurity to combat intensifying cyberthreats 10Geopolitics,export controls,semiconductors,advanced nodes,and AI 11Signposts for the future 12Whats inside2024 global semiconductor industry outlookAbout Deloittes OutlooksDeloittes 2024 global semiconductor industry outlook seeks to identify the strategic issues and opportunities for semiconductor companies and other parts of the semiconductor supply chain to consider in the coming year,including their impacts,key actions to take,and critical questions to ask.The goal is to help equip companies across the semiconductor supply chain with the information and foresight to better position themselves for a robust and resilient future.Chip sales look to bounce back in 2024,led by generative AI,but could be complicated by geopoliticsThe infamously cyclical semiconductor industry had a challenging year in 2023,the seventh downturn since 1990,with sales expected to be down 9.4%(to US$520 billion)for the year.1 But thats not as bad as was expected in the spring;before relatively stronger second and third quarters were in the books,the previous forecast had been for US$515 billion.2 2024 is now predicted to see global sales of US$588 billion.3 Not only would that be 13%better than 2023,but its 2.5%higher than 2022s record industry revenues of US$574 billion.The stock market is often a leading indicator of industry performance:As of mid-December 2023,the combined market capitalization of the top 10 global chip companies was US$3.4 trillion,up 74%from US$1.9 trillion in November 2022 and 17%higher than the US$2.9 trillion we saw in November 2021.4As is so often the case,the memory chip market was the biggest swing factor.In 2022,memory sales were almost US$130 billion,or just under 23%of the overall chip market,but they dropped 31%(about US$40 billion)in 2023,compared to down 1%for logic.The market is expected to get almost all of that back in 2024,with sales expected to reach 2022 levels.If we exclude memory,the rest of the industry was down in 2023,but only by about 3%.5In terms of end markets,both PC and smartphone sales are expected to grow 4%in 2024,after 2023 declines of 14%and 3.5%,respectively.6 Returning to growth for these two end markets is likely important for the semi industry:In 2022,communication and computer chip sales(which include data center chips)made up 56%of overall semiconductor sales for the year,compared to auto and industrial,which accounted for only 14%of sales each,for example.7Two other important measures of the industrys health are inventories and fab utilization.As of fall 2023,inventories remained high at more than US$60 billion,about the same level as the previous year.And the process of drawing those down will be a significant headwind for sales in the first half of 2024.8 In addition,utilization was high during the recent shortage(in the mid-90%range)and is expected to fall below 70%in Q4 2023.9 The industry likely needs utilization to be much higher than that to be profitable,which could take some time.Meanwhile,capacity is also growing as the United States and Europe increase domestic chipmaking.These trends and others play into our 2024 global semiconductor industry outlook,where we drill down into five big topics for the year ahead:generative AI accelerator chips and how semiconductor companies are using gen AI;trends around smart manufacturing;the need for more assembly and test capacity worldwide;how chip industry intellectual property(IP)is a target for cyberattacks at a whole new threat level;and a final geopolitics section that looks at export controls around advanced node manufacturing equipment and technologies,as well as advanced gen AI semiconductors.There are some interesting angles to the gen AI chip boom that affect the chip industry as a whole.With sales predicted to reach more than US$50 billion in 2024,10 this market is a tailwind for the sector and is expected to account for about 8.5%of sales.A portion of that will come from logic processors made on advanced nodes,some of it from advanced high bandwidth memory(HBM3),some from advanced 2.5D packaging,plus some from advanced connectivity chips.In each category,these gen AI-driven chips are among the priciest of their kind.In 2022,more than a trillion chips were sold at an average selling price of US$0.57 per chip.11 Meanwhile,some gen AI chips were selling for US$40,000 each in 2023,or 70,000 x higher,and therefore US$50 billion worth of chips might only be a volume of 1.25 million chips,or less than 0.1%of total chip volumes for the year.12Why does that matter?Although gen AI chips are expected to be a big part of 2024 chip revenues(and likely to be even bigger in the future),they are a relatively small part of unit volumes,and therefore manufacturing capacity.As mentioned above,with industry utilization below 70%,having gen AI chips fly off the shelves is great news for the handful of companies that sell those chips or parts of those chips,but it may not be as helpful for the overall industry.For the industry to see optimal utilization across all process nodes,other kinds of chips in addition to gen AI may need to see stronger demand.Executive summary2024 global semiconductor industry outlook32024 global semiconductor industry outlook4 4Selling gen AI chips:The market for chips that accelerate the training and inference of generative AI models was the semiconductor story of 2023.At a high level,gen AI chips are packages of special GPUs,special CPUs,special HBM3 in advanced 2.5D packaging,plus other special chips needed for connectivity in the data center.13 For an industry fighting headwinds from weak memory prices,and weak demand for smartphone and computer chips,gen AI chips provided a growth area,especially at leading manufacturing nodes.As 2024 approaches,the market for these chips looks to be strong and is predicted to reach more than US$50 billion in sales for the year,14 or 8.5%of the value of all chips expected to be sold for the year.15In the longer term,there are forecasts suggesting that AI chips could reach US$400 billion in sales by 2027.16 But what will happen in 2024?On the one hand,Deloitte predicted in November 2023 that gen AI chip sales will be more than US$50 billion in 2024.17 On the other,there are reasons to believe quarterly gen AI chip sales growth could flatten or even decline at some point,at least for a while.Fall 2023 was a perfect storm:strong demand,buyers eager to secure supply,and relatively few choices.But this could change at some point in 2024.1.New entrants:New chips are expected from existing gen AI chipmakers as well as emerging gen AI chipmakers.And new chips will likely come from companies that,up until now,have been better known as buyers of chips but have decided to start making their own.Nobody knows how any of these will fare at this point.2.New architectures and models:There are many different gen AI models and approaches,and no single chip architecture is likely to be optimal for all use cases.There will likely be data center chips,edge chips,training chips,and inferencing chips,and its expected that billions of dollars will be spent developing these“flavors”of gen AI chips.3.Edge chips:Its possible that in 2024 more processing will be done at the edge,on smaller,cheaper,or different accelerators or gen AI models than the ones that dominated in 2023.4.The bullwhip effect:The chip industry can be highly susceptible to over-ordering and excess inventories in times of undersupply,which then needs to unwind when supply catches up to demand,often leading to falling unit sales and falling prices per unit.This bullwhip effect18 could be seen by the second half of 2024.5.Will people pay for gen AI?Finally,much of the current demand for gen AI chips is coming from enterprise software companies either directly for those building on-premise processing,or indirectly via cloud companies that provide the software companies with gen AI processing services.They may plan on including gen AI features and services inside their existing or new software offerings and hope to charge for those features:Deloitte predicts this will provide a US$10 billion revenue uplift in enterprise software revenues by the end of 2024.19 And they may be buying up gen AI chips now(either directly or via the cloud)to help meet anticipated processing demand.Should software buyers be unwilling to pay a premium for gen AI services,or even be slow about doing so,the software companies could abruptly reduce their orders for gen AI processing.Generative AI and the next wave of chip supply 2024 global semiconductor industry outlook5Using gen AI for making chips:The chip industry has been using AI tools to help design chips for a couple of years,but that was just the beginning.20 Gen AI can help improve operations and proliferate best practices throughout the semiconductor industry value chain:According to Deloittes upcoming Gen AI in Semiconductors Study(2024),72%of respondents believe that gen AIs impact on their industry will be“high to transformative.”21 Respondents narrowed that further,describing the following areas where gen AI might be used,in addition to chip design and code creation:Generating more accurate schedules and supply chain forecasts Enhancing research and development via research augmentation Improving anomaly and defect detection;existing AI solutions are already at use in the industry,but gen AI offers the potential to speed this up by creating synthetic data for model training Operations could benefit from manufacturing process simulation and gen AI-enabled digital twin generation that can simulate sorting,assembly,testing,and other complex manufacturing processes without the need for petabytes of data Enhancing sales and marketing efforts by using gen AI for better content generation,tailoring marketing content to audience and purposeIn spite of these benefits,gen AI still has its challenges.Building or buying custom models for generating circuit design,testing plans,and synthetic data can be cost prohibitive as compared to manual execution.Running costs for very large models can also outweigh the speed benefits of automation.As is well known,gen AI can create inaccurate or nonsensical outputs,so validation by humans-in-the-loop is needed to improve accuracy.Finally,when working with human-centered applications such as HR or sales and marketing,human review can help ensure that all data is sanitized,personally identifiable information(PII)removed,guardrails installed,and validation performed.As gen AI chipmaking capacity increases,will prices and volumes continue to increase or come down?At a forecasted US$400 billion in 2027,gen AI chips would be close to half the chip industry;what would that mean?What early signals should be paid attention to from across the chip supply chain,such as movements in channel inventories or order buildups at hyperscalers?Many chip segments are dominated by a single player,whose preeminent role is both strong benchmark performance and a set of tools and support systems that lock buyers in.Will gen AI chips follow that trend,or will we see a more fragmented industry?To be prepared for the various scenarios,what changes will need to be made in supplier and vendor contracts,as well as order pipeline?Will there be aggressive M&A or private equity interest in smaller chip companies?Gen AI tools may transform the industry,but what will the effects be?Will they just lower costs,or will we actually see revenue growth?Strategic questions to consider:2024 global semiconductor industry outlook6Over the years,semiconductor fab facilities and outsourced semiconductor assembly and test facilities(OSATs)have leveraged IoT devices,robotics tech,and artificial intelligence/machine learning(AI/ML)and analytics with the goal of achieving smart,lights-out chip factories that are fully automated.22 Wafer fab equipment makers,integrated device manufacturers(IDMs),foundries,and back-end AT facilities all continue to invest more in smart manufacturing practices,digital tools,and technologies,but the basic smart manufacturing objectives more or less remain the same.They connect things on the factory floor,automate material movement and data collection,and apply analytics to prompt decisions and actions.However,from a chip manufacturing perspective,two things are expected to be different for smart manufacturing in the semi industry in 2024.One is the availability of sophisticated and highly advanced AI tools(including gen AI)to analyze large datasets and offer sharp insights.Deloittes upcoming Gen AI in Semiconductors Study noted roughly seven in 10 semi execs acknowledge gen AI will have a“high to transformative”impact on their business,indicating they view it as more than just another new tool.23 Across manufacturing,operations,and maintenance functions,semi execs surveyed believe gen AI has the potential to add the greatest value through analyses and insights pertaining to processes and equipment(noted by 28%of respondents),followed by predictive maintenance and smart diagnostics and troubleshooting(18%).The second relates to enhancing performance and sustainability of fabs and buildings.Moving from a mature technology node(e.g.,28 nm manufacturing)to an advanced node(2 nm)needs 3.5 times as much energy,consumes 2.3 times as much water,and emits 2.5 times as much greenhouse gases.24 The semi industry should consider implementing manufacturing transformation on older plants(brownfields),in addition to building brand-new greenfield plants,to help realize even greater sustainability benefits.25 Moreover,fabs can look at investing in smart manufacturing tools such as 6D BIM(building information modeling)to help improve cost management,simulate and analyze energy consumption,enhance efficiency,and streamline facility managementmaking a positive impact on the bottom line and the planet.26As fabs explore using smart manufacturing tools and virtual models such as 6D BIM throughout 2024,they should identify specific environmental and social factors and metrics to be measured for a more holistic sustainability assessment and environmental,social,and governance(ESG)reporting.Additionally,companies should consider exploring where and how technologies and tools such as gen AI,private 5G networks,and digital twins could be integrated in the manufacturing plants.27 Making smart manufacturing smarter 2024 global semiconductor industry outlook7Semiconductor executives polled as part of the Deloitte 2023 Semiconductor Transformation Study identified that lack of enterprise-wide alignment on digital capabilities is a key challenge as part of their transformation initiatives.28 Though the level of business and IT alignment showed improvement