Semiconductors have outperformed every other sector in technology this year, and I believe this is for good reason. Our site has been stated (consistently) for many years that the semiconductor industry is at the forefront of AI – and this is different than previous tech booms that were led by software. Although it’ll be a mix of hardware and software that drives the next leg up for artificial intelligence, we are firmly in the hardware stage of the hype cycle to where the competitive advantage for both data centers and enterprises is concentrated on the hardware, for now. In turn, this need for leading-edge hardware is contributing to high capital intensity.

Lam serves the memory market, and the cyclicality in memory is steep. DRAM is expected to decline (-39.4%) in 2023. Yet in 2024, DRAM will grow by “as much as 87 percent” according to Gartner. NAND flash memory is projected to decline by (-32.9%) and then “bounce back to grow by about 60 percent” in 2024. For semiconductor manufacturing equipment, the market is expected to decline (-22%) in 2023 and then rebound 21% in 2024. This is a remarkable turnaround, but one that should be approached carefully, as the rebound also represents a remarkable oversupply.

Ideally, we will enter Lam much lower than where it’s currently trading. The memory rebound is expected to ramp in June of 2024, so there is time. Micron is also showing evidence it will heavily participate in this rebound, although Lam is (typically) lower risk due to being diversified across the top 3 memory customers. As a reminder, Micron is #3 behind Samsung and SK Hynix, whereas Lam is tied for #1 on WFE with Tokyo Electron. The company also does quite well at disrupting the wafer fabrication equipment market with strong R&D spend and a rapid product cycle.

Over the past decade or so, Lam was considered lower risk because it was expected that memory manufacturers would continue to buy from Lam even during a low point in the cycle. This happened in 2015, when Lam was insulated from the last deep memory trough. However, due to the China ban, Lam did not escape the memory trough this time around.

The fundamentals are pointing toward either a bottom already occurring last quarter, or the bottom occurring this quarter. Fundamentals can always change – which is why we cover each earnings report for our core positions twice. However, as of now, the early part of 2024 is looking strong for Lam. We review these important key metrics below, and will ultimately wait for technicals to align before we consider an entry.

Lam Research Overview

You can access a previous deep dive here from 2020 and also a follow up in 2021 here.deep dive here from 2020 and also a follow up in 2021 here.

Water fabrication equipment (WFE) is a primary segment for Lam Research, specifically for memory and storage chips. The deposition process creates layers of insulating and conducting materials with techniques like chemical vapor deposition or atomic layer deposition, which allows for thin films of atomic layers to be coated onto surfaces. The excess material is then etched away.

Deposition and etch are processes that require complex machines for wafers to be built into integrated circuits.

3D NAND moves memory from a 2D plane to a 3D plane, which dramatically improves the storage capacity. Storage was reaching the end of its physical limits yet applications were requiring higher density and lower cost-per-bit. Due to data hungry applications – today in AI but also elsewhere, current NAND designs have been reaching the limits of monolithic die-level maximum capacity.

In 2013, Lam was at the forefront of this change by providing the equipment to manufacture 3D NAND. The advanced deposition and etch systems solved important manufacturing challenges for stacking film layers. Planar NAND has aspect ratios of 1:10 whereas 3D NAND has aspect ratios of 1:40. Going vertical makes it hard to maintain a uniform hole diameter.

Even more challenging is the need to fill narrow, horizontal features with a lateral deposition process. This is especially important for replacement-gate designs, which have replacement layers in the stack. We’ve detailed these differences in a Micron write-up here.

Cryogenic etch is a technology introduced a few years back where the etch removes the material in devices at cold temperatures below 100 degrees Celsius for high-aspect ratios with 200+ layers. The cold temperatures are achieved with liquefied nitrogen gas. You can click here for an article that describes this process.

There is competition in this space for Lam, especially from Tokyo Electron, who released a third generation cryogenic etch solution that lowers the temperature to increase the gas, which in turn, improves dimensions and depth over conventional cryogenic etch. This leads to a 53% reduction in etch time and is enabling 300-layer single-stack NAND by etching a 25% longer hole.

Lam has many highly technical processes for deposition and etch, here are a few. I won’t go into the details of each one, but rather have detailed the products most likely to drive forward growth below (advanced packaging, HBM/DRAM, and gate all around).

Kiyo and Flex process modules for the Sense.i platform, which increases output with a smaller warehouse footprint.
Coronus plasma bevel clean systems and Corvus etch for leading-edge node manufacturing
The company also supports Gallium Nitride manufacturing with “atomic level precision” and is porting it’s 300mm etch solutions to a 200mm process.
Additionally, advanced packaging is a segment where generative AI applications may drive more revenue for Lam. The company’s SABRE 3D copper electroplating and Syndion etch system is one of Lam’s solutions used for packaging advanced 3D stacking.

Lam’s Reliant Equipment is driving growth by providing a lower cost of ownership for non-traditional chip markets, such as micro electromechnical systems (MEMS), power chips, radio frequency (RF) filters and CMOS image sensors for better connectivity and more powerful imaging. This particular segment refers to trailing edge nodes, which means larger nodes, such as 24nm, 28nm or 90nm processes. Although we have written quite a bit about leading edge nodes, Lam has also found success in supplying equipment for larger nodes as these are used in automotive and medical equipment.

Lastly, Lam is a leader in plasma simulation, which allows engineers to model and simulate wafer builds to drive down costs and increase production time compared to the traditional method of building many wafer tests. In this case, engineers are testing virtual wafers, and this is expected to be popular for advanced nodes. Recently, Lam announced an initiative to train 60,000 engineers in India on its simulation software, Semiverse.

HBM/3D-Stacked DRAM

HBM and/or 3D-stacked DRAM and/or 3D DRAM are terms often used interchangeably, and are strong drivers for Lam given the company has an opportunity to supply the market with greater etch and deposition intensity to reverse DRAM’s slowing cost/bit declines.

The impetus is AI acceleration, which is in a stage where it’s fiercely competing on memory (see below excerpts from our AMD deep dive). In fact, according to Lam, AI servers use 8X DRAM and 3X NAND compared to an enterprise class server.

High bandwidth memory (HBM) offers higher bandwidth and lower power by vertically stacking memory chips to shorten how far data has to travel, while also allowing for smaller form factors. Stacked memory chips are connected through something called “through silicon vias” or TSVs.

TSVs are spoken about on Lam’s earnings calls, per the most recent comment: “we haven't exactly sized this business externally, but we — it is growing and, it’s in certain areas, especially as I highlighted things like etch for TSVs, copper plating for filling of those TSVs, pretty much anything related to the building of that advanced packaging structure through etch and deposition.”

Nvidia’s H100 utilizes HBM3 memory, supplied by both SK Hynix and soon Samsung will also supply the H100s. AMD’s MI300s are leverage HBM3 memory and 2.5D packaging, supplied by Samsung. Lam states the company has 50% market share in deposition and etch solutions in 3D stacking for high bandwidth memory.

The focus of our recent AMD deep dive in July helped emphasize how memory has become a point of fierce competition among AI accelerators. Here is what was stated in our report – I’ve bolded what’s important for this report on Lam:

“According to AMD, the MI300X will have 2.4X the memory density of the H100 and 1.6X the memory bandwidth. The reason that the MI300X was able to run the popular Falcon-40B large language model (LLM) with 40 parameters is because the neural network was ran entirely in memory without the need to move data back-and-forth with the external memory. AMD also stated the MI300X will be able to run up to 80B parameters on a single chip.

(8) 5nm GPUs and (4) 6nm base dies
153 billion transistors
192GB of HMB3 memory. 5TB/second of memory bandwidth
Ran a 40B parameter large language model (LLM) on a single GPU (unprecedented)
Can scale up to 8 accelerators in a single package for cutting-edge generative AI LLMs

The MI300X requires more power than its predecessor MI250X at 750 watts, and this is higher than Nvidia’s H100 at 700 watts. However, it’s not an apples-to-apples because what the MI300X promises to deliver is running compute-intensive large language models with fewer GPUs than is required with the H100s due to offering roughly double the memory.”

This month, Nvidia announced the GH200 super chip, which combines the H100 GPU with Grace 72-core Arm CPUs for an increase of memory capacity by 3.5X and memory bandwidth by 3X. Per the press release: “HBM3e memory, which is 50% faster than current HBM3, delivers a total of 10TB/sec of combined bandwidth, allowing the new platform to run models 3.5x larger than the previous version, while improving performance with 3x faster memory bandwidth.” This new generation of HBM3 memory is likely being supplied by SK Hynix.

Point being, memory is becoming an important component in the AI arms race, and Lam is a major equipment supplier and beneficiary of capital intensity that follows each new generation of HBM. Although Tokyo Electron is a notable and equal-sized competitor, Lam’s not to be underestimated on its ability to innovate quickly and serve the growing demands of the insatiable high-performance computing market.

Advanced Packaging

HBM3 and HBM3e require advanced 2.5D CoWoS packaging. Lam serves this market, and it’s expected to be an important growth driver over time. Per Lam’s management team: “I mean when you look at the importance of things like CoWoS-2 and AI packaged system, our exposure there even broadens further to the types of tools that we sell in there.”

Especially for AI purposes, chiplets are replacing monolithic circuits, where the overall system is divided into smaller parts so that 3nm or 5nm nodes can be replaced when needed. This avoids having to replace all of the components nodes shrink and design companies otherwise battle Moore’s Law.

With chiplets, AI platforms can scale by adding computing power and reduce total cost of ownership. Here is a whitepaper that discusses how chiplets are the way forward “Chiplet Cloud: Building AI Supercomputers for Serving Large Generative Language Models.” This whitepaper lays out some important arguments about the future of designs for AI acceleration. In most all cases, chiplets will be the desired architecture, and companies like Lam will provide these advanced packaging solutions.

Last year, Lam acquired SEMSYSCO to further its advanced packaging offerings for leading-edge logic chips and chiplet-based technologies. These solutions are needed for high-performance computing, AI and data-intensive applications. By cutting the chips or chiplets from a substrate sheet, chipmakers can increase yield and reduce waste.

Here are some of the stats shared on the most recent earnings call about advanced packaging:

“Customers are increasingly adopting a wide variety of packaging schemes to enable logic and memory integration. Some of these schemes enable up to 50% improved memory density, 10 times improvement in bandwidth and 60% gain in power efficiency. Lam has a track record of excellence in the advanced packaging segment with a strong set of manufacturing proven products.”

Gate All Around (GAA) Transistors

Lam supplies equipment for gate-all-around transistors, which are replacing replace FinFET transistors. Chipmakers such as TSM, Samsung and Intel produce GAA transistor designs.

FinFET transistors have a vertical geometry that helps to reduce the effects of Moore’s Law. This leads to more current flow with less leakage, leading to a lower gate voltage for operating the transistor. FinFET has been popular over the past decade as it resulted in better performance and lower power consumption.

Due to advanced nodes, FinFETS are reaching their limits in both height and how many can be stacked side-by-side. The vertical geometry is being replaced with horizontal nanosheets. This allows the gate to surround the channel on all four sides, reducing leakage and current. The result is better chip performance from the electrical signal that passes through the transistors.

A Few Reminders:

Given Lam’s highly technical product line, we should review the pros and cons for Lam so that the driving forces for this company does not get lost in the jargon.

There is a glut of supply in both NAND and DRAM, but especially NAND that must become absorbed. As you are all aware, the weak consumer market of PCs and mobile are weighing on otherwise strong companies (Apple, Microsoft, AMD, Broadcom, you name it). Memory is deeply feeling the effects of a weak consumer market. As a result, there is low production and low utilization rates in the industry right now, which will later lead to a shortage.

Lam’s advantage is the company’s ability to innovate at a fast clip to keep its equipment in high demand. Therefore, one could argue that even though memory has become commoditized, the equipment market can be leveraged within this competition for improved processes. The company’s customers include TSM, Intel, Samsung, SK Hynix and Micron. By serving the top 3 companies in memory, Lam has greatly outperformed the more well-known memory stock, Micron, over the past 10 years.

Lam’s production facilities are located in the United States, which is a positive as the CHIPS Act will benefit Lam. On the flip side, Lam is also exposed to the China ban, which has hurt Lam this year, and strengthened its competitors. The puts and takes to the geographical AI arms race have to be considered with Lam – and for all semiconductor companies really incl NVDA, AMD, MRVL — as this will only intensify as time goes on.

A Note on AI-related Revenue

The CEO of Lam stated that for “every incremental 1% penetration of AI servers and data centers is expected to drive $1B to $1.5B of additional WFE investment.” This is exciting but an analyst asked management to double click on this statement, as in this case, that 1:1 or 1:1.5 should be showing up by now given the surge in demand for AI servers and data centers.

This was an important part of the call, so I’ve included the transcript for you:

Atif Malik

“[…] When I listened to the major foundry in Taiwan, they talked about 50% AI semi growth rate in the next few years but we did not really raise any CapEx this year. And we believe AI server adoption is probably like high single digit, 8%, 9%.

So what explains the discrepancy in terms of not seeing a lift in leading-edge investments from the higher server adoption? Maybe it's just underutilization in other end markets.

Doug Bettinger

I guess what I think about, Atif, is these are longer-term statements we're making. I think you're right about the composition of AI servers. In the short term, you're not really going to see a meaningful uptick in WFE, frankly. This is going to occur over the next several years. As more of — and you're right, I think server volumes largely, I don't know, mid- to, I'd call it, mid-single digit percent of total servers. I think that grows over time. But in the short term, it doesn't really show up as quickly as you might think. It requires sort of capital investments to occur for future demand. That doesn't really happen in a meaningful way in the short term.”

Financials

Lam Research saw as high as 49% revenue growth in 2021 when we owned the stock and will now bottom at (-32.7%) growth in the upcoming September quarter. Assuming analyst estimates are correct, Lam will then rebound to 20% growth by the June 2024 quarter, and will remain at this growth rate into the foreseeable future.

This rebound is seen across a few semiconductors with consumer exposure, such as Taiwan Semi and Micron. Lam has an exceptional bottom line, and this is a large part of our consideration. As stated in the intro, fiscal year EPS is eye-watering at $27.05 expected for FY2024 and will grow to $55.13 EPS by FY2028. This is not a typo with the decimal point.

Analysts were positive on Lam’s gross margins as they bottomed in the March quarter at 41.5% and are now at 45% and 46%.

Even during the downturn, Lam maintained a strong GAAP operating margin of 26.6% last quarter, this is down about 500 basis points from its peak. The low on operating margin is likely behind us, at a 24.4% margin in the March quarter. Current guide for this upcoming quarter is an operating margin of 27%.

Similar to operating margins, cash flow margins have maintained quite well. Operating cash flow margin last quarter was 35% and free cash flow margin was 32.5% for a little over $1 billion in FCF. The company has $5B in cash and $5B in debt on the balance sheet. This has been consistent for many quarters.

Gross margins reached a 5 and 10 year low in March 2023. Operating margins reached 5-year lows and you can see where Lam improved operational efficiency.

Meanwhile, based on company September quarterly guidance. Quarterly sales also appeared to have bottomed out in the June 2023 quarter.

Consensus has revised eps mainly due to better-than-expected profitability from self-help and bottoming in sales. Lam has also been buying back shares to lift EPS.

Key Segments:

To drive forward growth, all three segments from Lam are important to monitor. Foundry is where we will see growth from leading-edge devices and Logic/other is where advanced packaging revenue will be recognized. However, for a true recovery, we need Memory and DRAM to bottom as it’s a substantial part of Lam’s business. You can see the blue arrow below is pointing toward Memory/DRAM declining from 40% of revenue and 14% of revenue, respectively, to 18% and 9% of revenue in the most recent quarter.

Combined Memory was at 27%, which per management, is the “lowest concentration percentage for this segment in the last decade.” Management also stated they expect “NAND spending to remain at low levels for the remainder of the 2023 calendar year” and is the lowest they’ve seen since the 3D NAND entered the market circa 2013.

Given Lam’s strong performance over the past decade, coupled with its strong bottom line and penchant for innovation, a 10-year trough may be a buying opportunity – provided we can get the stock at the right price.

When Will Lam Bottom?

Canary in the coal mine

Given the financial metrics we’ve outlined, most point to signs Lam is getting closer to the bottom in the “cycle”. The section below answers what will we be focusing on to give us confidence that the 2024 growth rates that Gartner has estimated are realistic?

Lam breaks down their sales between two business units – Systems and Customer Support Business Group (CSBG). The Systems business sells the hardware. CSBG is the aftermarket business that provides services and parts to its clients after they’ve installed Lam’s hardware.

This is a recent breakdown in sales between these two businesses. We’ve estimated the September quarter breakdown based on management’s guidance of $3.4b in total sales.

During periods of strong demand, the breakdown between the two is about 65%/35% between Systems and CSBG. This reflects strong sales of new hardware and the steady growth of the aftermarket that follows. During periods of weaker demand, the percentage of hardware sales decreases and aftermarket increases because there are fewer new systems sales and the aftermarket is primarily done on existing capacity.

As can be seen above, this ratio peaked at 67/33 in the December 2022 quarter and since then has steadily declined to 53/47 in the June 2023 quarter, which is primarily a function of the decline in Systems sales. The 47% CSBG contribution is the highest ever.

Looking ahead we see signs we’re closer to the bottom. Firstly, based on Lam’s September guidance, we believe that the June quarter marked the bottom in Systems sales and that September will show sequential positive growth. Secondly, we believe CSBG will bottom in the September quarter if not the December quarter.

The focus on when the CSBG segment has bottomed is important because it is an early indicator that Lam’s clients are planning to increase utilization. This is how Lam described its importance when asked in the recent June quarterly call as to what investors should focus on for signs that the environment is getting better.

Pricing

An important catalyst for an increase in utilization and capex will be higher DRAM and NAND prices. A chart by the Yole Group shows the recent history and estimated Q/Q price changes. Lam’s fab clients will begin to increase utilization if Q/Q pricing begins to improve.

In the most recent call, Lam indicated that NAND capex has declined the most and management has heard of recent, additional cuts: “And again, just listening to our customers and talking to them, we've heard some customers talk about further cuts in NAND just in the last 24 hours.” Management also stated that DRAM will be the first to recover as they are optimistic around HBM.

China exposure

Lam’s China exposure is both a blessing and a curse. The former in that it comprises 26% percent or $4.5b of Lam’s FY2023 sales. The latter in that it is at center of the export restrictions targets that the US government has aimed at China in the name of IP security. In the most recent call, Lam stated that they had lost about $2b in sales due to these restrictions. In response, Lam has focused on areas that aren’t subject to these restrictions. The risk is that these restrictions will spread to those areas and will continue to meaningfully impact Lam’s China business. This risk is quantifiable. Although highly unlikely, in the worst-case scenario Lam could lose 26% of its sales overnight.

From June 2021 to June 2022 to June 2023, Lam’s China revenue has gone from 37% to 31% to 26%. A reasonable base case is to assume this downward trend will continue over time.

Delay in Capex

If pricing does not improve, WFE capex may take longer to resume and WFE sales may take longer to resume.

Based on estimates from semi.org, 2023 WFE sales are expected to be about $75b. And if WFE sales have bottomed, Lam 5.1 p/sales (1 yr. fwrd) vs its 4.6 avg, is not yet pricing in a significant improvement.

Interestingly, despite the sharp decline in sales in q3 and q4 in FY2023, the FY2023 annual sales were flat y/y. Lam expects a slight pick-up from China and HBM. If Systems sales are beginning to improve and CSBG is close to bottoming, consensus estimates of 14% decline in Lam’s FY2024 sales appears a bit pessimistic.

In the most recent June quarter, deferred revenue was $1.8B vs $2.B in March. Regarding the deferred revenue, Lam made the following comment.

“We've got all the deferred revenue related to outstanding back order parts completely back to normal levels. As we discussed last quarter though, our deferred revenue balance is currently still at higher than historic levels.”

VALUATION:

Due to the tech rally, which was led by semiconductors this year, Lam Research is trading at a 1-year high and 3-year high of 5.9 forward P/S ratio. The PE Ratio is the same, it’s trading at a 1-year and 3-year high of 24X.

TA will be important as the stock is close to 2022 all-time highs. Lam tested first support level at $644 and bounced to $662. Next support is at $587.

Advanced Signals Members receive real-time trade alerts for our entries and in-depth technical analysis from the Portfolio Manager, Knox Ridley. Learn more here.Learn more here.

Conclusion:

The pummeled consumer market is deceiving as eventually hybrid AI will bring AI processing capabilities to the edge, including consumer devices. Memory is a key component in the competitive AI race both inside and outside the data center.

Our goal is to identify an AI-related memory stock by mid-2024 that we can own into the foreseeable future. We’ve tracked Big Tech capex as a proxy for AI acceleration in the data center, and we are forming new proxies now for the consumer device AI boom. Apple will always be Apple, but their margins will likely be lower with semiconductor companies benefiting from higher dollar chip content per device. This will take time to play out, but if we are lucky, then the consumer weakness will depress the very stocks that will most benefit from AI moving toward the edge.

Lam Research: Wafer Fab Equipment Leader & HBM/DRAM Memory