HBM/CXL Emerge in Response to Demand for Optimized Hardware Used in AI-driven HPC Applications, Says TrendForce

According to TrendForce’s latest report on the server industry, not only have emerging applications in recent years accelerated the pace of AI and HPC development, but the complexity of models built from machine learning applications and inferences that involve increasingly sophisticated calculations has also undergone a corresponding growth as well, resulting in more data to be processed. While users are confronted with an ever-growing volume of data along with constraints placed by existing hardware, they must make tradeoffs among performance, memory capacity, latency, and cost. HBM (High Bandwidth Memory) and CXL (Compute Express Link) have thus emerged in response to the aforementioned conundrum. In terms of functionality, HBM is a new type of DRAM that addresses more diverse and complex computational needs via its high I/O speeds, whereas CXL is an interconnect standard that allows different processors, or xPUs, to more easily share the same memory resources.

HBM breaks through bandwidth limitations of traditional DRAM solutions through vertical stacking of DRAM dies

Memory suppliers developed HBM in order to be free from the previous bandwidth constraints posed by traditional memory solutions. Regarding memory architecture, HBM consists of a base logic die with DRAM dies vertically stacked on top of the logic die. The 3D-stacked DRAM dies are interconnected with TSV and microbumps, thereby enabling HBM’s high-bandwidth design. The mainstream HBM memory stacks involve four or eight DRAM die layers, which are referred to as “4-hi” or “8-hi”, respectively. Notably, the latest HBM product currently in mass production is HBM2e. This generation of HBM contains four or eight layers of 16Gb DRAM dies, resulting in a memory capacity of 8GB or 16GB per single HBM stack, respectively, with a bandwidth of 410-460GB/s. Samples of the next generation of HBM products, named HBM3, have already been submitted to relevant organizations for validation, and these products will likely enter mass production in 2022.

TrendForce’s investigations indicate that HBM comprises less than 1% of total DRAM bit demand for 2021 primarily because of two reasons. First, the vast majority of consumer applications have yet to adopt HBM due to cost considerations. Second, the server industry allocates less than 1% of its hardware to AI applications; more specifically, servers that are equipped with AI accelerators account for less than 1% of all servers currently in use, not to mention the fact that most AI accelerators still use GDDR5(x) and GDDR6 memories, as opposed to HBM, to support their data processing needs.

Although HBM currently remains in the developmental phase, as applications become increasingly reliant on AI usage (more precise AI needs to be supported by more complex models), computing hardware will then require the integration of HBM to operate these applications effectively. In particular, FPGA and ASIC represent the two hardware categories that are most closely related to AI development, with Intel’s Stratix and Agilex-M as well as Xilinx’s Versal HBM being examples of FPGA with onboard HBM. Regarding ASIC, on the other hand, most CSPs are gradually adopting their own self-designed ASICs, such Google’s TPU, Tencent’s Enflame DTU, and Baidu’s Kunlun – all of which are equipped with HBM – for AI deployments. In addition, Intel will also release a high-end version of its Sapphire Rapids server CPU equipped with HBM by the end of 2022. Taking these developments into account, TrendForce believes that an increasing number of HBM applications will emerge going forward due to HBM’s critical role in overcoming hardware-related bottlenecks in AI development.

A new memory standard born out of demand from high-speed computing, CXL will be more effective in integrating resources of whole system

Evolved from PCIe Gen5, CXL is a memory standard that provides high-speed and low-latency interconnections between the CPU and other accelerators such as the GPU and FPGA. It enables memory virtualization so that different devices can share the same memory pool, thereby raising the performance of a whole computer system while reducing its cost. Hence, CXL can effectively deal with the heavy workloads related to AI and HPC applications.

CXL is just one of several interconnection technologies that feature memory sharing. Other examples that are also in the market include NVLink from NVIDIA and Gen-Z from AMD and Xilinx. Their existence is an indication that the major ICT vendors are increasingly attentive to the integration of various resources within a computer system. TrendForce currently believes that CXL will come out on top in the competition mainly because it is introduced and promoted by Intel, which has an enormous advantage with respect to the market share for CPUs. With Intel’s support in the area of processors, CXL advocates and hardware providers that back the standard will be effective in organizing themselves into a supply chain for the related solutions. The major ICT companies that have in turn joined the CXL Consortium include AMD, ARM, NVIDIA, Google, Microsoft, Facebook (Meta), Alibaba, and Dell. All in all, CXL appears to be the most favored among memory protocols.

The consolidation of memory resources among the CPU and other devices can reduce communication latency and boost the computing performance needed for AI and HPC applications. For this reason, Intel will provide CXL support for its next-generation server CPU Sapphire Rapids. Likewise, memory suppliers have also incorporated CXL support into their respective product roadmaps. Samsung has announced that it will be launching CXL-supported DDR5 DRAM modules that will further expand server memory capacity so as to meet the enormous resource demand of AI computing. There is also a chance that CXL support will be extended to NAND Flash solutions in the future, thus benefiting the development of both types of memory products.

Synergy between HBM and CXL will contribute significantly to AI development; their visibility will increase across different applications starting in 2023

TrendForce believes that the market penetration rate of CXL will rise going forward as this interface standard is built into more and more CPUs. Also, the combination of HBM and CXL will be increasingly visible in the future hardware designs of AI servers. In the case of HBM, it will contribute to a further ramp-up of data processing speed by increasing the memory bandwidth of the CPU or the accelerator. As for CXL, it will enable high-speed interconnections among CPU and other devices. By working together, HBM and CXL will raise computing power and thereby expedite the development of AI applications.

The latest advances in memory pooling and sharing will help overcome the current hardware bottlenecks in the designs of different AI models and continue the trend of more sophisticated architectures. TrendForce anticipates that the adoption rate of CXL-supported Sapphire Rapids processors will reach a certain level, and memory suppliers will also have put their HBM3 products and their CXL-supported DRAM and SSD products into mass production. Hence, examples of HBM-CXL synergy in different applications will become increasingly visible from 2023 onward.

For more information on reports and market data from TrendForce’s Department of Semiconductor Research, please click here, or email Ms. Latte Chung from the Sales Department at


NEV Market Remains Bullish Across 1Q21-3Q21 Period Against Headwinds, with More Than Four Million Vehicles Sold Globally, Says TrendForce

Total global sales of NEVs (new energy vehicles) for the first three quarters of 2021 (January-September) reached 4.2 million units, with BEVs in particular accounting for 2.92 million units, a 153% YoY growth, according to TrendForce’s latest investigations. Total sales of PHEVs, on the other hand, reached 1.28 million units, a 135% YoY growth. Compared to the overall automotive market, whose growth has been constrained by the ongoing semiconductor shortage and effects of the COVID-19 pandemic, sales of NEVs still remained relatively strong.

Regarding BEV sales, Tesla comfortably took the leadership position with a 21.5% market share. The automaker’s sales volume for the first three quarters of this year already surpassed its sales volume for 2020. Taking second place on the top 10 list, Wuling Hongguang was able to maintain its high volume of sales due to not only low retail prices, but also a gradual expansion of its target markets from tier-three and tier-four cities to tier-one and tier-two cities in China. This shift would seem to indicate a corresponding expansion of and shift in Wuling Hongguang’s customer base. BYD and Volkswagen took third and fourth places, respectively, with the latter aggressively consolidating its BEV offerings into the ID. Family this year. Vehicles in the ID. Family have accounted for nearly all of Volkswagen’s BEV sales since 3Q21. Despite the rapid growth of the BEV market, competition has been intensifying after traditional automakers began releasing their own BEV models at a faster pace while emerging automakers also began delivering vehicles.

It should be noted that, although the global semiconductor shortage has not damaged the NEV market to the same degree as it did the traditional ICE vehicle market, the NEV market is not entirely immune to the resultant supply-side issues. In addition, China’s power rationing and pandemic-generated transportation/logistics disruptions likewise affected automakers’ manufacturing operations to various degrees. Taken together, these aforementioned factors became some of the underlying causes responsible for the shifts in NEV automakers’ market shares.

Regarding PHEV sales, BYD put up a remarkable performance by leapfrogging to second place in the rankings, and this can primarily be attributed to the release of BYD’s DM-i vehicles, which feature a super hybrid technology aimed at reducing fuel consumption. Thanks to the DM-i vehicles, BYD’s PHEV sales began skyrocketing in 2Q21, and the automaker was able to overtake several European automakers with respect to total PHEV sales for the first nine months of 2021. Much like the BEV market, despite the growths in most automakers’ sales volumes, companies will find it increasingly difficult to raise their PHEV market share.

Looking ahead to the NEV market’s future, TrendForce believes that, as traditional global automakers gradually kick off mass production of vehicles based on the battery electric platform, more and more new BEV models will be released to market at an accelerated pace. Furthermore, the next one to three years will serve as the key timeframe for emerging automakers as well as new entrants that crossed from other industries to achieve mass production. Therefore, there remains much potential for changes to occur within the rankings of NEV automakers’ sales and market shares.

For additional insights from TrendForce analysts on the latest tech industry news, trends, and forecasts, please visit our blog at


Shortage of Semiconductor Parts, Such As IC Substrates, Becomes Primary Driving Force Behind Development of FOPLP Technology

As the COVID-19 pandemic wreaked havoc on the global electronics supply chain, the packaging and testing operations of mid-range and high-end chips were subsequently confronted with prolonged lead times. This can primarily be attributed to the fact that IC substrate suppliers were unable to raise output or expand their production capacities in the short run in order to meet the skyrocketing volume of client orders. Hence, products that are packaged using BGA (Ball Grid Array), Flip Chip, or SiP technologies, all of which require the use of IC substrates, had their lead times lengthened. Certain IC design companies are therefore considering the feasibility of packaging technologies that do not require substrates.

Regarding the trend of advanced packaging development, technologies such as 2.5D/3D IC, SiP, and FOPLP (Fan-out Panel Level Packaging) remain the current mainstream R&D targets. Given the ongoing shortage of semiconductor components, including IC substrates, FOPLP, in particular, has garnered the most attention among the aforementioned three packaging technologies as it can be operated without substrates. At present, most OSAT companies and other chipmakers have successively invested in FOPLP-related technological and manufacturing development in order to capitalize on potential new commercial opportunities.

Despite FOPLP’s advantage of packaging chips across large areas, technological development remains problematic

Regarding the history of FOPLP development as well as the technology’s evolution going forward, its earliest roots can be traced to existing packaging applications including Flip Chip and BGA. As end-products continued to experience performance upgrades, leading to the number of I/O pins being insufficient for meeting the increase in performance demand, new types of wafer-level packaging technologies such as Fan-in and Fan-out subsequently emerged to fulfill the packaging demands of mid-range chips, high-end chips, and other emergent applications.

Although Fan-in and Fan-out packaging technologies are able to effectively raise the number of I/O pins, they also result in a substantial increase in manufacturing costs compared to previous-generation technologies such as Flip Chip and BGA. For both 8-inch wafers and 12-inch wafers, packaging costs have only been on a very slight downtrend. That is why the packaging industry has placed a top priority on simultaneously lowering production costs while raising the number of chips packaged at once. Hence, FOPLP technology has emerged in response to this demand for large-area packaging technology.

Regarding the actual implementation of FOPLP, a potential solution may be found in wafer-level packaging RDL (Redistribution Layer) designs, such as chip first or chip last. It should be noted that chip first FOWLP or chip last FOWLP processes do in fact serve as feasible concepts for FOPLP development. However, the FOPLP process involves stacking massive amounts of packaging materials and chips together, and their combined weight may lead to such issues as panel warpage. In addition, it remains difficult to maintain a consistent uniformity and yield rate during the FOPLP process, meaning further collaborations and optimizations on the parts of OSAT companies and semiconductor equipment suppliers are necessary for FOPLP to succeed going forward.

(Image credit: Unsplash)


Annual Automotive MLCC Demand for 2021 Expected to Reach 449 Billion Pcs, with Potential for Additional Growth Next Year, as Suppliers Compete Over Shares in Automotive Market, Says TrendForce

Various MLCC suppliers’ book-to-bill ratios as well as quarterly shipments for 4Q21 now show signs of decline, according to TrendForce’s latest investigations. Not only has the demand for consumer electronics slowed, but ODMs’ clients have also eased their procurement activities due to issues including the global chip shortage, mismatched component availabilities, and China’s power rationing. Demand in the automotive market, on the other hand, has remained strong since 3Q21. Automotive applications have therefore become an important point of focus in MLCC suppliers’ latest product planning and capacity expansion efforts. Thanks to these in-demand applications, annual MLCC demand from the automotive market for 2021 is expected to reach 449 billion pcs, a 20% YoY increase.

TrendForce further indicates that the growth of the EV market and improvements in ADAS specifications have resulted in a twofold increase in automotive MLCC consumption. While EVs’ electrified drivetrain and high safety requirements represent a high barrier to entry for MLCC suppliers, these hurdles have also in turn raised MLCC products’ ASP and profitability. Hence, the automotive electronics industry has been increasing its annual MLCC demand by double-digits in recent years. In particular, an analysis of different vehicles and their respective MLCC consumption reveals the following: a conventional EV requires 2.2 times the MLCC usage of a conventional gasoline vehicle, an ADAS-equipped EV requires 2.7 times, and an autonomous EV requires as much as 3.3 times.

Regarding MLCC suppliers, Japanese companies including Murata, TDK, and Taiyo Yuden continue to dominate the automotive MLCC market. These suppliers will expand their production capacities for automotive applications in overseas facilities in China, Philippines, and Malaysia next year, with powertrains, ADAS, and connected systems being among the most significant of the aforementioned applications. Korea-based Samsung, on the other hand, specializes in powertrain applications by leveraging its MLCC offerings’ small form factor, high capacitance, and high voltage. Finally, Taiwanese suppliers, such as Yageo and Walsin, are actively invested in developing automotive products and High-Q products for RF applications in an effort to increase their presence in the infotainment system market and EV charging station market.

Looking ahead to 2022, TrendForce expects annual automotive MLCC demand to reach 562 billion pcs, a 25% YoY increase, primarily attributed to the continued electrification of vehicles. While the global implementation of carbon-neutral policies and excellent sales performances of Tesla vehicles bring about widespread adoption of EVs, various countries have successively set concrete dates for the termination of gasoline vehicle sales. Hence, EVs are gradually becoming not only the mainstream option in the automotive market, but also the primary driving force behind the future growth of the MLCC industry.

For more information on reports and market data from TrendForce’s Department of Semiconductor Research, please click here, or email Ms. Latte Chung from the Sales Department at


Annual DRAM Revenue for 2022 Expected to Reach US$91.5 Billion, with Prices Likely to Rally in 2H22, Says TrendForce

Despite the forecasted 18.6% YoY growth in total DRAM bit supply next year, the global DRAM market is still expected to shift from a shortage situation to an oversupply, according to TrendForce’s latest investigations. This shift can primarily be attributed to the fact that, not only are most buyers now carrying a relatively high level of DRAM inventory, but DRAM bit demand is also expected to increase by only 17.1% YoY in 2022. On the price front, the oversupply situation will result in a drop in DRAM ASP in 2022 but not a major decline in annual DRAM revenue, thanks to the oligopolistic nature of the DRAM industry. Annual DRAM revenue for 2022 is expected to reach US$91.54 billion, which represents a slight YoY increase of 0.3%.

Based on an analysis of DRAM sufficiency ratio (which refers to the surplus of supply in comparison with demand) for each quarter in 2022, TrendForce forecasts a 15% YoY decrease in DRAM ASP for 2022, with prices undergoing the more noticeable declines during the first half of the year. Heading into 2H22, however, owing to the rise in DDR5 penetration rate, as well as the arrival of peak seasonal demand, the decline in DRAM ASP will likely narrow. TrendForce does not rule out the possibility that DRAM ASP may even hold flat or undergo an increase in 2H22.

Annual NAND Flash revenue is expected to experience yet another increase next year by 7.4% YoY while numerous suppliers compete in higher-layer NAND Flash market segment

Turning to the NAND Flash market, TrendForce forecasts a 31.8% increase in total bit supply for 2022 and a 30.8% increase in total bit demand. Hence, NAND Flash ASP will likely experience a downtrend next year as a result of the oversupply situation. In addition, due to the perfect competition in the NAND Flash market, the decline in NAND Flash ASP next year will be more noticeable than the decline in DRAM ASP. However, NAND Flash suppliers continues to make progress in the stacking of NAND Flash layers, so the growth in NAND Flash bit supply next year will therefore remain above 30%. TrendForce thus expects NAND Flash revenue to have more room for growth and reach US$74.19 billion in 2022, a 7.4% YoY increase.

TrendForce’s forecast based on an analysis of NAND Flash sufficiency ratios for each quarter in 2022 similarly points to an 18.0% YoY decline in NAND Flash ASP next year. Much like DRAM, NAND Flash prices will undergo the more noticeable declines during 1H22. Arrival of peak seasonal demand in 2H22 will potentially result in a narrowing of price drops and a potential for quarterly prices to hold flat.

On the whole, the revenue performance of the DRAM industry and that of the NAND Flash industry over the years show that the annual total DRAM revenue is growing at more stable pace. Again, this has to do with the oligopolistic structure of the DRAM market. Since the DRAM market has a different competitive landscape, the fluctuations in the overall DRAM ASP have been relatively modest over the long run. However, the development of the DRAM manufacturing technology is approaching a physical bottleneck as process nodes shrink below the 20nm level. This means that the bit growth derived from the deployment of a more advanced process is becoming more and more limited over the years. On the other hand, not only are NAND Flash suppliers relatively more unstable in their capacity expansion plans compared to the DRAM industry, but further improvements in NAND Flash layer-stacking technology also remains feasible. Hence, the fluctuations in the overall NAND Flash ASP have been relatively more volatile over the long run. On account of these factors, the DRAM industry generally has smaller YoY revenue growth rates compared with the NAND Flash industry, although the DRAM industry continues to surpass the NAND Flash industry in terms of profitability.

Profitability of suppliers may be constrained if total revenue fails to keep pace with continuously rising CAPEX

Regarding the CAPEX (capital expenditures) of DRAM suppliers, there has been a gradual increase in these suppliers’ CAPEX to sales ratio in recent years, for two reasons. First, the development of the DRAM manufacturing technology is approaching a physical bottleneck. Die improvements have become more and more limited after process nodes have shrunk below the 20nm level. Micron’s 1alpha nm process can offer an almost 30% increase in bits per wafer, but the 1Xnm-to-1Ynm migrations and the subsequent 1Ynm-to-1Znm migrations that the major suppliers have undertaken in the recent period have yielded increases of no more than 15% in bits per wafer. Looking at future technological developments, Samsung and SK hynix have already integrated EUV lithography into their most advanced process technologies. However, orders for EUV lithography tools have a much longer lead time, and the costs of these tools are also high. Hence, the three dominant suppliers have allocated a large chunk of capital expenditure in advance to place orders for EUV lithography tools ahead of time.

Secondly, the oligopolistic structure of the DRAM market has also helped establish a regime where there is a very low chance of a supplier’s ASP dipping under its fully-loaded cost despite the recurrence of the cyclical price downturn. Moreover, DRAM suppliers have accumulated a substantial amount of profit from their products. In view of the difficulties in die shrinking, suppliers ranging from the three dominant suppliers to others with less market share (such as Nanya Tech and Winbond) have developed tangible plans for capacity expansions. These plans have, in turn, become the other main driver behind the ongoing increase in the CAPEX to sales ratio.

The CAPEX to sales ratio of NAND Flash suppliers have likewise risen substantially following the transition to 3D NAND technology in 2017. Notably, the average CAPEX to sales ratio fell within the 25-30% range prior to 2017, but it has since climbed to nearly 40% as of now. This growth can primarily be attributed to the fact that, as the number of 3D NAND layers increases, there is a corresponding increase in the lead times of NAND Flash products and in the degree of precision as well as difficulty involved in the etching process. While the mainstream layer count of NAND Flash products approaches 1YY layers, suppliers are currently planning to move forward with the development of products with 2XX layers, which place an ever-increasing demand on etch depth. The CAPEX of NAND Flash suppliers will continue to grow alongside increases in layer count and revenue.

TrendForce indicates that NAND Flash layer-stacking technology will continue to progress, meaning suppliers will continue to pursue the stacking of additional layers as a way to lower their manufacturing cost per GB. As such, the NAND Flash industry’s CAPEX will have additional potential for growth going forward, with a CAPEX to sales ratio of close to 40% or above. It should be noted, however, that if total NAND Flash revenue fails to keep pace with the growth in CAPEX in the next few years, NAND Flash suppliers’ CAPEX to sales ratio may potentially undergo an excessive increase, thereby constraining the profitability of suppliers.

For more information on reports and market data from TrendForce’s Department of Semiconductor Research, please click here, or email Ms. Latte Chung from the Sales Department at

  • Page 4
  • 32 page(s)
  • 159 result(s)