Articles
See All
Aehr Test Systems – From EV’s to AI
AEHR TEST SYSTEMS (AEHR – NASDAQ)
FROM EV’S TO AI
The stock of AEHR Test Systems (AEHR-NASD) has done a full-circle run from $7 – $51 back down to $10 in the last five years, as its chip-testing technology went through the semi-conductor investment cycle.
It could be on the cusp of another big run. There is A) a big short position and B) when they report next week (Jul 16), they could hint or even say that they are about to find some BIG customers in the AI datacenter world.
That could potentially rock this beaten down small cap NASD stock—after all, it would now have a unique technology in TWO big global markets—Electric Vehicles and AI datacenters.
Investing Whisperer subscribers remember Aehr as we owned the stock from October 2022 until February 2023. While we sold out too early (isn’t that always the case), we still cashed in a 50% gain from $22 to $33.
The stock kept going. And going. And going. Aehr peaked out at $50+ in the summer of 2023.
Since then, Aehr has come right back down. Today, Aehr is down 80% from the top.
In the last 9 months Aehr’s business has slumped. They reduced their FY 2024 revenue guidance twice in two months – from $100M to $80M and then to $65M. Revenue in their fiscal Q3 (which ends in March) was $7.6 million, down 56% from $17.2 million in Q3 last year
What caused the slide? Aehr is EXTRA vulnerable to the semiconductor equipment cycle.
Aehr makes test and burn-in equipment. Their products are part of the line of machines used to make semi-conductors. Purchases of capital equipment are lumpy at the best of times.
Semiconductor equipment is a concentrated industry. Aehr is a small player in a land of giants like KLA Corporation (KLAC – NASDAQ), Applied Materials (AMAT – NASDAQ) and ASML (ASML – NASDAQ). They rely on just a few big customers for most of their orders. Their top-5 accounts made up virtually all their sales. Their top-2 customers accounted for 79% of sales last year.
Second, Aehr’s biggest customer has been ON Semiconductor (ON – NASDAQ). ON uses Aehr equipment to test their inverter components that they sell to electric vehicle manufacturers.
We all know the EV market is hurting. Sales are not as strong as forecasts.
ON is feeling the pinch and has reduced staff and guidance themselves.
It’s worse for Aehr. As a provider of capital equipment, Aehr not only needs EV demand to increase, but it needs to increase enough to justify more capital equipment purchases.
If ON sees flat sales, they don’t need more equipment. Which is what has happened over the last 6-9 months.
While EV demand will eventually turn, there are no signs of that just yet. Once it does, there will be a lag for when new capital equipment orders will be needed.
So does that make Aehr’s stock dead weight? Not so fast. EV’s aren’t Aehr’s story this time. Instead, Aehr is making inroads in other end markets: silicon photonics, AI accelerators and the big cahuna – MEMORY.
There are hints that the turn in some of these markets is coming. ON Semi announced in mid-June that they “planned a multi-year investment of up to $2B in a silicon carbide manufacturing facility in the Czech Republic.
The facility will be used to produce chips not only for EVs, but for renewable energy and AI data center applications.
Some of that cash will most likely be used to purchase Aehr equipment.
While revenue from these markets may still be quarters away, any inkling of success and investors will anticipate what’s coming – markets like memory and data centers are just that big.
Yet the stock is depressed, and many investors are betting against Aehr. Perhaps because the stock was such a high-flyer, Aehr has quite a high short interest. According to Capital IQ, there is 20%+ of the shares short.
Aehr reports earnings on July 16th. CEO Gayn Erickson loves to drop hints on their quarterly calls. A few well-placed tidbits about engagements around data centers and AI could send the stock into a squeeze.
NUTS AND BOLTS
The decision to use Aehr equipment comes down to a trade off that every semiconductor manufacturer weighs – reliability vs. cost.
Aehr builds test and burn-in systems. These are large, expensive machines that they sell to semiconductor manufacturers. Aehr’s main product is the FOX-XP system.
Source: Aehr Investor Presentation
A semi-manufacturer uses the FOX-XP system to “burn-in” their products. Burn-in means exposing the semiconductor to high temperatures and pressures that simulate a year worth of stress in a few hours.
If a semiconductor is going to fail, it will most likely do it in the first year. By simulating the first year before selling the device, you can weed out the weak chips and increase reliability of the rest.
Source: Aehr Investor Presentation
What makes Aehr’s FOX-XP system unique is what it tests. Traditional test and burn-in equipment test the semiconductor at the chip level – a fully put together chip. Aehr’s FOX-XP tests at the wafer level, when the chip is a “sheet” of silicon dies that have not yet been assembled.
Source: 300mm NAND Wafer with between 200-450 die on it
Aehr’s uphill battle (for the last 10 years!) has been convincing semiconductor manufacturers to test their components at the wafer level and not the chip level, if they test at all.
That decision is all about cost vs. reliability. Chip manufacturers don’t want to spend more to produce their chips than they have to. If their customers are okay with the reliability, the manufacturer would prefer not to do any burn-in at all. If the customer says they want better reliability, the chip manufacturer starts to do burn-in.
But that burn-in has typically been at the chip level.
Aehr’s value is when chip level burn-in is no longer possible.
As chips get more complicated, in particular with the advent of modular chips that have multiple dies and components, failure becomes more costly.
If a modular chip has 10 components all packaged together, and you have to throw the whole thing away because one die has failed, that can be a BIG waste. Aehr solves that problem.
It is a problem that continues to grow as more chips become more complicated. Aehr’s time is coming.
SILICON CARBIDE –
THE ORIGINAL AEHR USE CASE
Aehr’s solution is not the cheapest option, which is why Aehr has only slowly made inroads into end markets.
Their first breakthrough was with EVs. The reason that Aehr made inroads into electric vehicles was because:
- EVs use a multi-device inverter to manage the EV drivetrain.
- The inverter requires EXTREMELY high reliability – otherwise the cars would just stop.
EVs use inverters to convert DC battery power into AC power that the electric motor can use.
Source: Aehr Investor Presentation
Inverters modules have many silicon carbide devices in them that do the work of converting power.
Testing the silicon-carbide devices before they are assembled into the inverter is right in Aehr’s wheelhouse. Instead of throwing away an inverter with 24 dies because one die failed, Aehr’s test system can catch those failures ahead of time.
Stepping back to see the forest through the trees, Aehr benefits from:
- Greater chip complexity
- Demand for better reliability.
These are two long-term trends that make the stock attractive whenever it sells off too much. Especially when we see green shoots across other end markets, which is what I see now.
MEMORY – THE HOLY GRAIL OF SEMICONDUCTOR CAPITAL EQUIPMENT
Years ago, when Aehr first designed their FOX-XP test and burn-in system, the anticipated market for the device was memory.
Flash forward 7-8 years and that hasn’t planned out – yet. But the desire to get their foot in the door of the memory market has never waned.
The memory market is huge. It is estimated to be around $150B worldwide. Memory is expected to make up over 50% of the total semiconductor wafer shipments worldwide in 2024.
For years Aehr has struggled to find a lead customer in memory. But that may be about to change.
Memory chips consist of many silicon dies that are stacked on top of one another. These stacks keep getting bigger. Micron is up to 232 layers with their NAND memory.
Testing these at the chip level mean more expensive failures the more layers that you have. Same story – if one die is bad, you throw out the whole chip. There’s a tipping point where the cost of failure outweighs the cost of Aehr.
The proliferation of systems on chip (SoC) devices also push the needle towards Aehr.
An SoC compresses all the components onto a single piece of silicon.
Source: Ansys SoC Example
SoC’s are becoming ubiquitous with AI because having memory, clocks, etc. very close to GPUs helps reduce time lags and GPU downtime.
The downside of SoC’s is that if you burn-in the entire module, one memory failure means you throw out the entire device.
Aehr CEO Gayn Erickson said that Aehr is “driving for our first on-wafer benchmark in partnership with a leading NAND supplier using our proprietary WaferPaks and FOX wafer level test and burn-in system”.
“We see the memory market as a significant opportunity for us to deliver wafer level burn-in solutions to help memory suppliers meet their reliability and quality needs, particularly with stack die applications”
High failure rates are becoming a big problem for AI. The datacenters that are being built to run the next-gen AI training are absolutely massive. They have 100,000s of GPUs, 10,000s of optical transceivers, millions of memory devices.
That is driving some of the memory use case and is also driving another datacenter component – optical transceivers.
REDUCING THE HIGH FAILURE RATE OF OPTICAL TRANSCEIVERS
Silicon Photonics is the type of chip used in optical transceivers. These are devices used to covert electrical signals from circuit boards to optical signals that can be sent via fiber wires.
Optical transceivers are becoming more important in AI datacenters that require faster speeds – something that fiber, which can go the speed of light (at least theoretically) can deliver.
The problem is A. optical transceivers have high failure rates and B. the scale of these datacenters means they have as many as 10,000-40,000 transceivers!
Consider that if an optical transceiver has an average life of 5 years, and if you have 10,000 transceivers in your AI datacenter, the chance that one will fail in the first hour is extremely high.
Failures of transceivers can put a BIG wrench in the works. AI requires basically everything talking to everything – meaning the failure of a single transceiver can put the process on FULL STOP until it is replaced.
Gayn Erickson spoke of transceivers in the Aehr’s FQ2 call in February.
We remain very enthusiastic about this market, which includes the current photonics transceiver market used in data and telecommunications and the upcoming application of silicon photonics integrated circuits for use in optical chip-to-chip communication, which we see as a major market opportunity
Aehr’s FOX-XP “allows for the testing of as many as 8,000 high-power optical devices in parallel on each of 9 wafers before they’re singulated and placed into a photonic application”, thus weeding out weaker units and reducing failures.
ACCELERATOR CHIPS –
YET ANOTHER AI USE CASE
Finally, Aehr is getting their foot in the door with AI accelerators.
Part of this opportunity comes back to memory. Aehr is seeing interest in wafer testing where there is “co-packaging [of] memory in AI processors”.
AI accelerators are putting DRAM on the same silicon as the processor. Testing that memory before creating the module becomes necessary but so is testing the accelerator chips themselves.
These AI accelerator chips pack more functionality on the same piece of silicon to increase speed. As complexity goes up (more dies on the same silicon) so does potential failure.
Aehr sees these “multi-chip modules” as yet another opportunity. Their newest FOX-XP multi-wafer burn-in system was designed to handle “several thousand watts of power and over 2,000 amps are current” on 9 wafers, which is the high power you need to burn-in these devices.
BUY NOW AND WAIT?
That’s the good news – opportunities abound! The bad news? None of this is likely going to drive revenue growth for the next couple of quarters.
The next few quarters could be ugly. To get a sense of just how quickly Aehr’s EV related ON Semiconductor business hit the breaks, consider the following chart:
Source: Aehr Financial Statements
Revenue went from a run rate of $20M+ to less than $8M virtually overnight.
When Aehr’s business is doing well, they can generate a lot of FCF.
Source: Aehr Financial Statements
If Aehr had been able to maintain their original guidance for F24 of $100M+ they would have likely had even higher FCF margins. $30M FCF was not out of the question.
Unfortunately, that didn’t happen. The company is back to treading water at single digit $M quarterly revenue.
Aehr does have some recurring revenue. They sell WaferPaks for their Fox-XP each time a customer needs to transition to a new form factor of design. These wafer packs are actually expected to make up >50% of revenue in F24.
Of course, that is largely because Aehr isn’t selling many of their systems. The wafer pack business becomes more interesting as Aehr scales. If Aehr hits the motherlode with a big memory win you can imagine a time where the memory upgrade cycle drives large waferpak upgrades as well.
But for now, Aehr remains a story of selling test systems.
Aehr has enough opportunities that it is working through that I feel pretty confident in saying that they will be back to growth in the medium term. $80M+ annual revenue in a couple years is a fair base case. Whether that happens in 2025 or 2026, I am less certain.
That gives the stock significant upside, but not necessarily right away.
The wildcard is the short interest. There are some hints, such as the ON Semiconductor announcement I referenced earlier, that the capital equipment cycle is turning. A few well-placed comments could send that large, short interest in the stock covering, which could drive the stock up even in the absence of revenue.
As usual, you take your chances. You buy the stock here at $12 with the understanding it could go to $9 before it goes higher if there is more weakness in the business and/or the market finally takes a swoon.
Why buy it now? Because you just never know. When the inflection hit Aehr could move VERY fast. The upside is $20+ on a recovery in their silicon carbide market and potentially much higher if they get systems sold into the memory market.
