With the advancements in AIGC models such as ChatGPT and Midjourney, we are witnessing the rise of more super-sized language models, opening up new possibilities for High-Performance Computing (HPC) platforms.

According to TrendForce, by 2025, the global demand for computational resources in the AIGC industry – assuming 5 super-sized AIGC products equivalent to ChatGPT, 25 medium-sized AIGC products equivalent to Midjourney, and 80 small-sized AIGC products – would be approximately equivalent to 145,600 – 233,700 units of NVIDIA A100 GPUs. This highlights the significant impact of AIGC on computational requirements.

Additionally, the rapid development of supercomputing, 8K video streaming, and AR/VR will also lead to an increased workload on cloud computing systems. This calls for highly efficient computing platforms that can handle parallel processing of vast amounts of data.
However, a critical concern is whether hardware advancements can keep pace with the demands of these emerging applications.

HBM: The Fast Lane to High-Performance Computing

While the performance of core computing components like CPUs, GPUs, and ASICs has improved due to semiconductor advancements, their overall efficiency can be hindered by the limited bandwidth of DDR SDRAM.

For example, from 2014 to 2020, CPU performance increased over threefold, while DDR SDRAM bandwidth only doubled. Additionally, the pursuit of higher transmission performance through technologies like DDR5 or future DDR6 increases power consumption, posing long-term impacts on computing systems’ efficiency.

Recognizing this challenge, major chip manufacturers quickly turned their attention to new solutions. In 2013, AMD and SK Hynix made separate debuts with their pioneering products featuring High Bandwidth Memory (HBM), a revolutionary technology that allows for stacking on GPUs and effectively replacing GDDR SDRAM. It was recognized as an industry standard by JEDEC the same year.

In 2015, AMD introduced Fiji, the first high-end consumer GPU with integrated HBM, followed by NVIDIA’s release of P100, the first AI server GPU with HBM in 2016, marking the beginning of a new era for server GPU’s integration with HBM.

HBM’s rise as the mainstream technology sought after by key players can be attributed to its exceptional bandwidth and lower power consumption when compared to DDR SDRAM. For example, HBM3 delivers 15 times the bandwidth of DDR5 and can further increase the total bandwidth by adding more stacked dies. Additionally, at system level, HBM can effectively manage power consumption by replacing a portion of GDDR SDRAM or DDR SDRAM.

As computing power demands increase, HBM’s exceptional transmission efficiency unlocks the full potential of core computing components. Integrating HBM into server GPUs has become a prominent trend, propelling the global HBM market to grow at a compound annual rate of 40-45% from 2023 to 2025, according to TrendForce.

The Crucial Role of 2.5D Packaging

In the midst of this trend, the crucial role of 2.5D packaging technology in enabling such integration cannot be overlooked.

TSMC has been laying the groundwork for 2.5D packaging technology with CoWoS (Chip on Wafer on Substrate) since 2011. This technology enables the integration of logic chips on the same silicon interposer. The third-generation CoWoS technology, introduced in 2016, allowed the integration of logic chips with HBM and was adopted by NVIDIA for its P100 GPU.

With development in CoWoS technology, the interposer area has expanded, accommodating more stacked HBM dies. The 5th-generation CoWoS, launched in 2021, can integrate 8 HBM stacks and 2 core computing components. The upcoming 6th-generation CoWoS, expected in 2023, will support up to 12 HBM stacks, meeting the requirements of HBM3.

TSMC’s CoWoS platform has become the foundation for high-performance computing platforms. While other semiconductor leaders like Samsung, Intel, and ASE are also venturing into 2.5D packaging technology with HBM integration, we think TSMC is poised to be the biggest winner in this emerging field, considering its technological expertise, production capacity, and order capabilities.

In conclusion, the remarkable transmission efficiency of HBM, facilitated by the advancements in 2.5D packaging technologies, creates an exciting prospect for the seamless convergence of these innovations. The future holds immense potential for enhanced computing experiences.

NVIDIA’s robust financial report reveals the true impact of AI on the technology industry, particularly in the AI server supply chain. …

Apple’s latest MR device, the “Vision Pro,” utilizes Micro OLED technology. This technology, along with Micro LED, is considered the next generation of display technology. So what are the differences between Micro OLED and Micro LED, and which one is better suited for AR/VR/MR devices?

According to market research firm TrendForce, ideal smart glasses must meet three major criteria. Firstly, to minimize the burden of wearing glasses, the display engine’s size should be below 1 inch. Secondly, in terms of content recognition requirements, the display brightness specification should reach at least 4,000 nits to ensure immunity to external factors such as weather or venue conditions. Lastly, the resolution should be at least 3,000 PPI to ensure clear projection and magnification.

Currently, Micro LED and Micro OLED are the primary technologies that meet these requirements. However, Micro LED is still in the early stages of AR technology development and faces several challenges that need to be overcome. Therefore, Micro OLED is currently the mainstream technology in the field.

Micro OLED technology enables full-color capabilities and has become the preferred choice for AR/VR manufacturers. According to TrendForce’s comparison of display engines, Micro LED outperforms Micro OLED in pixel size, luminous efficiency, and brightness. It appears to be the most suitable for AR glasses based on specifications. However, Micro LED is currently limited to a single green color, while Micro OLED can achieve full color. As a result, Micro OLED has a competitive advantage in AR/VR devices.

In terms of manufacturers, Sony remains the main supplier for Micro OLED technology. Due to their longer investment time and technological advantages, South Korean manufacturers Samsung and LG Display (LGD) are expected to join Apple’s MR supply chain in 2024.

Last year, reports suggested that Samsung initially considered Micro OLED a niche market and lagged behind its competitor, LGD. However, due to demands from Apple, Meta, and Samsung’s parent company, they began developing Micro OLED in the third quarter of last year. The latest news reveals that Samsung will acquire American Micro OLED display manufacturer eMagin for a price of $218 million.

Meanwhile, Meta will also collaborate with South Korean semiconductor giants SK hynix and LGD to develop Micro OLED panels for Meta XR (Extended Reality) devices. This partnership is expected to lead to more Micro OLED applications in AR/VR in the future.

Micro LED technology is still facing bottlenecks, but it has the potential to surpass Micro OLED in the medium to long term. TrendForce states that Micro LED AR glasses, due to the bottleneck in achieving full colorization, primarily display monochromatic information such as informational prompts, navigation, translation, and note-taking functions. Achieving higher resolutions requires chip miniaturization, reducing the size of Micro LED to 5 micrometers. In this situation, epitaxial processes are affected by wavelength uniformity issues, which impact yield. Additionally, smaller chips raise concerns about the external quantum efficiency (EQE) of red chips.

Overall, although Micro LED faces many challenges in AR glasses, it still outperforms Micro OLED in contrast, responsiveness, lifespan, power consumption, and other specifications. Considering the limitations of waveguide component technology in transparent AR glasses, which restricts optical efficiency from exceeding 1%, Micro LED remains an excellent choice in the medium to long term.

Therefore, if Apple wants to introduce Micro LED technology, it plans to start with the Apple Watch. However, the project’s launch has been delayed from 2024 to a later date, possibly beyond 2025, due to technological bottlenecks. In fact, over the past decade, Apple has invested significant funds in collaboration with ams Osram to develop Micro LED components. Once the technology is ready for mass production, Apple is likely to take charge of the critical “mass transfer” process, which may be carried out at its secret research and development center in Longtan, Taoyuan.

It’s worth noting that in addition to Micro LED, the Longtan research and development center is also where Apple collaborates with TSMC on Micro OLED technology for MR devices.

(Photo credit: Apple)

Tesla, the driving force behind the next-generation electric vehicle(EV) battery standards, has been vigorously promoting the 46800 cylindrical battery cell in recent years.

Being Tesla’s key collaborator, Panasonic had initially scheduled mass production of these batteries for April this year. However, in a recent announcement, they revealed that their production plans would be delayed by at least a year, with full-scale production not set to kick off until between April and September 2024.

This strategic pivot is aimed at optimizing performance, but what we care about is the implications it might hold for the EV supply chain – could this mean that the strong alliance between these two giants is beginning to waver? And if so, what sort of ripple effect could this have on the relevant market?

Tesla’s secret weapon in the EV price war

Given the capacity of 46800 battery cell is five times that of the 21700 battery, it means fewer cells are required to achieve the same total battery pack capacity.

For instance, a 75kWh-based Model 3 uses 4,416 units of the 21700 battery cells packed in the traditional way of Cell to Module (CTM), which assembles batteries into modules which are then encased into a battery pack and then fitted onto the vehicle’s chassis.

In contrast, a Model Y with the same battery capacity would need only 828 units of the 46800 battery cells, leading to a 14% saving on battery costs. Coupled with Tesla’s integrated chassis technology (CTC), where batteries are not assembled into modules but instead directly encapsulated under the cabin floor, this provides an ultimate, cost-effective solution for Tesla.

When Tesla first announced its 46800 battery plan in 2020, its battery capacity was pioneering among all batteries. Taking advantage of this favorable environment, Tesla has been both expanding their production and involving cylindrical battery manufacturers, like Panasonic, in their comprehensive plans.

Tesla has set up a 46800 battery production line at their Fremont factory in California. As of the end of 2022, their production capacity was about 4GWh, which can only support 50,000 to 60,000 75kWh EVs and is far from their sales volume.

In terms of a long-term strategy, Tesla not only aims to ramp up their production capacity but is also heavily reliant on external suppliers like Panasonic to support its ever-growing demand.

Hence, ever since the launch of Model S in 2012, Panasonic has remained Tesla’s primary supplier of power batteries. And thanks to Tesla’s booming sales, Panasonic has dominated the power battery market for quite a while.

Roadblocks for Tesla and Panasonic’s Alliance

So, what does Panasonic’s delay mean for its position in the market? In fact, as an important chess piece in Tesla’s battery market strategy, Panasonic has been under considerable pressure.

Externally, there’s the relentless price cuts from Tesla. In 2018, as Tesla’s sales skyrocketed, they started purchasing batteries from more suppliers, thereby indirectly pressuring Panasonic to lower prices.

In addition, the internal discord has also been shadowing the project. On one hand, the long-term supply to Tesla has not brought as impressive profit performance as anticipated for Panasonic’s battery business. On the other hand, sticking to Tesla’s technology route, Panasonic has missed a great deal of opportunities to partner with Japanese car makers due to its conservative investments in the mainstream hydrogen energy batteries, which has in turn stirred internal questioning.

Since 2020, both South LGES and CATL have become suppliers to Tesla, causing Panasonic’s market share to fall to third place globally. But even then, Panasonic’s many years of expertise in cylindrical batteries made it Tesla’s Top choice when deciding to manufacture the 46800 battery. This was widely seen as Panasonic’s best chance to regain its leading ground and to solidify long-term partnership with Tesla.

Is Panasonic about to miss out on its prime opportunity?

All in all, we believe that this delay could not only disrupt Tesla’s price war strategy but also make Panasonic miss the golden chance to secure its dominance in the new technology. With multiple battery manufacturers, such as CATL, LGES, and Eve Energy, announcing that they will start mass production of the 46800 battery in 2024 or 2025, Panasonic will face unprecedented competition.

As of Q1 2023, Panasonic has seen its market share fall to fourth place. Obviously, maintaining its industry leadership becomes more of a daunting task for the company in the race. Although they’ve announced plans to build at least two 46800 battery factories in North America, it won’t serve as a panacea for their problems.

Beyond overcoming technical hurdles and expediting mass production, Panasonic also has a mountain to climb in terms of diversifying its customer base, further lessening the risk of an over-reliance on Tesla. These are inevitably long-term challenges that Panasonic cannot sidestep.

In the bustling tech bazaar, the iPhone 14 Pro and AirPods 3 are pioneering the tech industry by incorporating InP(Indium phosphide)-based EEL(Edge Emitting Laser). These devices are leveraging the unique attributes of long-wavelength technology for skin detection, which is a strategic move that highlights the gradual emergence of InP material in the consumer market.

Historically, data communication and telecom industries have acted as the primary fuel for the InP market, their demand for backbone network photoelectric and 400G/800G optical modules in data centers has been consistent. However, as the quality and refinement of 6-inch InP single-crystal growth technology advance, we see a reduction in production costs, thus unlocking a gateway to consumer applications.

Emerging Dual Frontiers: Consumer and Photonic Applications

Apple and other savvy smartphone OEMs are contemplating the introduction of long-wavelength InP-based EEL in their next-gen products. This would be used for physiological sensing in proximity sensors or possibly to replace the currently used 940nm GaAs-based VCSEL(Vertical Surface Emitting Laser) in 3D sensing.

Simultaneously, the evolution of autonomous driving is nudging automotive laser radars towards the 1550nm wavelength, a departure from the former 905nm. This shift promises increased detection range and improved protection for human eyes.

In the realm of photonics communication technology, a more significant growth driver stems from the trend of high-end EML(Electro-absorption Modulated Laser) replacing traditional DFBs(Distributed-feedback laser).

As next-gen data center applications are steered towards 400G/800G transmission speed solutions, EML laser chips promising high bandwidth performance and high yield will take the spotlight. They are anticipated to realize the high-speed transmission characteristics of single-wavelength 100G.

It is also worth noting that as fiber-optic access in the PON (Passive Optical Network) market gradually upgrades to the 25G/50G-PON solution, there is an evident trend towards integrated solutions combining laser chips and SOAs (Semiconductor Optical Amplifiers). This shift is driven by the increasing demands for higher transmission rates and output power, leading to the replacement of discrete DFB solutions.

Supply Chain Over-centralization: A Precursor to a Sellers’ Market?

This cornucopia of application scenarios signals tremendous market potential for InP-based components. However, one must question whether the supply chain is prepared for this windfall.

One of the concern is that the industry chain’s over-centralization might usher in a seller’s market situation.

InP substrate materials and epitaxial silicon wafers pose a high technological threshold and are primarily monopolized by few manufacturers, particularly those from Europe, the U.S. and Japan.

• The InP substrate material market is highly monopolized by Sumitomo Electric Industries, AXT, and JX NMM, which collectively account for 90% of market share in 2020.
• The epitaxy process is the crux of photonic chip production, with tech prowess directly impacting product performance and reliability. Key suppliers capable of providing InP epitaxy silicon wafers include IQE, Lumentum, and Sumitomo, among others.
• In terms of photonic chip technology, its value lies more in added functionality, necessitating process integration. This gives rise to IDM giants dominating the market. For instance, Lumentum, Sumitomo, and Mitsubishi dominate the 25G DFB laser chip market.

While the influx of newcomers from China is seen in the lower-tech optical module packaging sector, the core technologies upstream are still held firmly by international industry leaders, posing a challenging breakthrough for newcomers in the short term.

The growing interest in the market for this technology indicates that end-product manufacturers developing new applications based on InP will inevitably need to double down their efforts to ensure the stability of long-term supply. It remains to be seen whether the singularity of the supply chain will further restrict the proliferation of emerging applications in the end market.