Aehr--Memory Angle?

This source claims each FOX-XP system is $2.5 million, not $25 million.

Adding the automation increases cost another $1m (almost) if you’re doing all the trays, which would seem to be the norm.

Consumables aren’t really consumables. You don’t need a new WaferPak until your design changes. Those happen every few years.

Some envelope math:
18 wafers max per FOX-XP.
1,000 chips per wafer max.
Assume 6 hours per burn-in test (varies).
That’s 72,000 chips/day output.
At Tesla’s old 48 chips/inverter, that’s 1500 cars/day. Tesla builds almost 5,000 cars/day world-wide, so you’d need 4 systems to support Tesla’s world-wide output in 2023. 50% more next year, unless Tesla is successful in reducing SiC usage per vehicle.

13 Likes

The burn in time tends to be more like 12 hours. Some go 24 hours. And some, like silicon photonics go 72 hours.

Also, even with automation there is down time between batches and machines seldom run 24/7 as they require adjustment, maintenance, and monitoring.

So to cover Tesla will probably take more than 4 machines and growing as Tesla expands. STM is their primary provider with ON as their other supplier. Don’t know how the two compare volume wise these days between them. Plus SiC is used in more than the inverter and a certain percentage of wafers are burned out. Each wafer can be cut into many chips. So the math is not so simple.

But does bring up the case why STM is not a larger AEHR customer.

6 Likes

AEHR’s net revenue last quarter was $20.6m. If we take the $2.5m machine, add $1.5m for the first set of WaferPaks and add just under $1m for automation aligners, we’re looking at $5m/machine in revenue. Given AEHR’s other products, that’s just a 2 or 3 FOX-XPs sold last quarter. Unless my math/numbers are wrong.

So, whether it’s 4 machines or even 10 machines, to cover Tesla this year, with Tesla selling almost 2 million vehicles out of about 10 million EVs total (and hard to know what percentage of those use SiC, maybe less than half) this year, how many more machines can AEHR be selling? Yeah, revenue could double during the next year, but isn’t that priced into the stock price already? Can AEHR double revenue again?

Remember, the “razor/blade” model doesn’t actually apply to AEHR after all.

It would seem unlikely that a SiC producer would spend the money to test older designs they know they’re going to replace soon - that $1.5m WaferPak cost needs to test enough chips to be worth the money. So, companies like STM or ON might be waiting for their new designs to be further down the pike before buying and setting up test machinery. See my note in another thread on SiC architecture moving from Planar to Trench.

13 Likes

Good analysis. Tesla (drawing from Henry Ford’s original strategy) does stick with designs for a long time. Tesla is not redesigning their chips every year. Tesla is coming out with new models but I’d guess they reuse their chip designs as much as they can.

Tesla will be cresting redesigned cars and coming out with their pickup truck so those will likely have some redesigned chips and may be a catalyst. Speculation of course as I am only going from what seems likely given how Tesla operates. Still, all the old chip designs still need to be burn in tested. Perhaps STM is using some older method as STM was the first SiC supplier some years ago. Before AEHR really began being a thing.

Speculation but good analysis Smorgasbord.

7 Likes

@Smorgasbord1, I think you’re half right. Yesterday I got to thinking about what I wrote and figured that price tag for a FOX-XP with all the bells an whistles was just too high. I googled it . Here’s the source google provided:
" [### Aehr Receives $25.1 Million Order for FOX-XP Test and …
(Aehr Receives $25.1 Million Order for FOX-XP Test and Burn-in Systems to Support Production of Silicon Carbide Power Devices for Electric Vehicles)
But, I just went with the headline and failed to read the article. Oops. The article goes on to say a significant number of FOX-XP systems without quantifying significant. My bad.

But I said half right, SiC burn in takes minimum 12 hours. Some chip makers even go 24 hours. But if we just go with the minimum requirement that doubles your estimate to meet the requirement.

Further, I don’t know where you got 1,000 chips per wafer. I’ve searched a lot to find the chip density for a 200mm SiC wafer without any luck. I do think 1,000 chips per 200mm wafer for Si memory chips might be about right. But I think it’s a big leap to assume that SiC chips used in inverters has the same density. In addition, if they got 100% yield as you suggest they wouldn’t need to do burn in testing at all. The infant mortality rate for SiC chips is a bit more than 1%. While that doesn’t change your math a lot, it does reduce the throughput.

Another thing I think you got wrong was the 48 chip count. The number of SiC chips used is model dependent. Model S Plaid has three motors. Each motor has an inverter so that would be 144 SiC chips. The Cybertruck has four motors, hence four inverters. That implies 192 SiC chips. OK, that’s not really significant this year, but we’re talking about the “near future”.

I know, Musk recently said (I paraphrase): “I think that we might be able to reduce the number of SiC chips we use in the model S some time in the future.” There’s a lot of qualification in that statement. Musk is usually right about what he says he can deliver, but he’s also usually overly optimistic about when he delivers it.

Another thing you failed to consider is that during the most recent CC, 1Q24, Gayne (CEO) said they were confident that they will have revenues this year in excess of $100M. In the 4Q23 conference call Gayne all but admitted that was a sandbag number. So if they can be very confident of revenue over $100M this year, which is about 100% over FY23, do you really think another 100% over this year can’t be reached any time in the near future? I think some time in the near future is inevitable. The demand for SiC chips is ramping. Applications beyond automotive are already in development and to a limited extent, production. True, that doesn’t represent much SiC chip demand today, But again, the “near future” is the time frame.

Smorg, you’re smart and knowledgeable. I respect your thinking. But, on this one I’m pretty confident that you got it wrong.

9 Likes

@brittlerock you are correct, you don’t get 1000 chips from even a 200 mm SiC wafer. The number is more like 258 plus a few perhaps to that on a perfect wafer. Defects along the edge of the wafer are most common. A 200 mm wafer has less “edge” area per surface area than does 150mm SiC chip.

Also, yes burn in time is at least 12 hours for SiC and 24 hours is not out of the question. Burn in time for silicon photonics is expected to be longer than this and at higher power levels.

The 800 volt battery makes a real difference as burning in and orphan testing SiC requires higher power levels and at these levels chip damage from traditional equipment occurs at as little as 900 volts when the testing needs to be 1200 colts or higher. AEHR has this capability. Thus, why 800 volt batteries are such an important thing for AEHR’s customers as we move into 24, 25, 26.

The earnings transcript talks about many of these things. The new thing I did not like was ‘25, ‘26 are when 800 volt new models really begin to take off. In my mind '24 is when the S curve begins to steepen further. Thus why AEHR has built out its inventory and production capacity as it has been meet this projected SiC demand, which is way more than current product demand. One would still think '24 sees a steepening of the S curve with things going mainstream in 25 and 26.

11 Likes

Sure. In another thread I said it could be 10 instead of 4. Heck, it could even be 20. What number for the EV gorilla in the room would give you warm and fuzzies about AEHR’s potential growth?

No, each Tesla inverter has 24 SiC chips. Since, until recently, all Model 3 and Y’s were dual motored, that’s where I got the 48 count. We can ignore Model S Plaid as its volumes are so low as to not matter. Remember, Tesla sold only 16,000 Models S & X combined last quarter. I’d guess less than 10% are Plaids, and even 8,000 Plaids a year makes up for the RWD (1 motor) Models 3 & Y. As for CyberTruck, we do not yet know if Tesla will even make a quad-motored version. Tesla gave up on “Plaid Plus” after all.

No, the comments on SiC were from Colin Campbell, VP of Powertrain. They were accompanied with photos and slides. Here’s a link to the video starting with his Campbell’s comments on inverters, though I would suggest watching from a few minutes earlier:

He starts off with what they’ve “already been able to do.” First up is that Tesla has designed custom packages that extract twice as much heat from the SiC chips as they could buy off the shelf. Campbell: “This means that the Silicon Carbide wafer that’s inside those packages can be much smaller and silicon carbide, it’s an amazing semiconductor but it’s also expensive and really hard to scale. So using less of it is a big win for us.”

This is what Tesla is already doing. Now, let’s get to the future. Start at 58:42 in the video:

Campbell: “In our next powertrain, the silicon carbide transistors that I mentioned are a key component but expensive, we’ve figured out a way to use 75% less without compromising the performance or efficiency of the car.”

Note there is no “I think.” There is no “some time in the future.” This is stuff they’ve already figured out and it’s coming in the next powertrain. Here’s the slide:

Note that the $1,000 price is for the entire drive unit - that’s including the motor, not just the inverter.

No, this was a big part of my calculations. After all, revenue last quarter (Q1) was $20.6m, so they only need to beat that by 25% during the next 3 quarters to hit that $100m number. And when you consider that a fully fleshed out FOX-XP is about $5m, we see that sales are extremely low volume. 25 FOX-XP all year could do that.

I’m not sure what your “near future” timeframe is. AEHR just completed Q1 of their fiscal year, so “over this year” means at the earliest the next fiscal year, and frankly, that’s just too far out for me right now.

One of my concerns is that sales are predicated on movement of high priced, low volume units that have long sales cycles. CEO Erickson admitted as much on the recent ER call:

Candidly, this is the longest and most extensive sales and benchmarking process I’ve ever experienced in my entire 30-plus year career.

Is AEHR in denial of what Tesla is doing? From https://www.power-and-beyond.com/tesla-announces-that-it-will-use-75-fewer-sic-transistors-in-its-powertrains-a-2a7e01b32fb2a8647b75bdb0d2c22fad/

Aehr said new chips used in lower-cost Tesla models will have double the amps per device and likely will be 50 % or larger in surface area. “Therefore, the number of wafers required will be less impacted.”

But we already heard Colin Campbell of Tesla say they’ve cut the surface area needed by half due to superior heat transfer, AND that they’re going to reduce another 75% in the next powertrain. Has AEHR talked about the impact of Trench architecture on SiC chip production and volumes? I couldn’t find a peep on the web.

Let’s face it, this will never be a high-volume business. And this business is unlike anything I can recall being discussed on this board before. There’s no subscription. There’s no recurring revenue. It’s the rear-end of a manufacturing test business that no-one wants to spend any money on if they don’t have to. And the CEO admits the number of potential customers remains small:

We don’t have hundreds of customers and never will in this space, but we’ll definitely have a much more diverse number of customers, which is nice.

So, what is here? It’s really just a narrative story about the demand for SiC chips increasing as more EVs are going to be made. Adjacent uses like electric trains and such are so low in volume as to not matter. Future things like GaN are probably half a decade out, unless Tesla is hiding something. When will silicon photonics testing take off - what chip volumes per year are we talking about, and what year?

For now, we have the player that created AEHR’s largest market in 2017, and remains the largest player in that market, saying they’re going to use 75% less SiC per vehicle in their next gen powertrain. The excuses given by others after that presentation (some of which I linked earlier) were that they weren’t counting on Tesla in their future predictions. Excuse me while I roll my eyes.

23 Likes

Erickson was quoting a UBS forecast that 30% of the $8B SiC market will be for 800 volt automotive systems by 2026. That would seem to require that Tesla also migrate to an 800 volt system. Here’s an article on 800 volt systems in EVs:

Automakers Porsche, Hyundai, Genesis, Kia and Audi already offer EVs with 800-volt battery systems. Volvo, Polestar and Lotus have also committed to 800-volt architectures. Hitachi Automotive Systems is starting mass production of its 800-volt battery system.

and on Tesla:

Interestingly, EV pioneer Tesla has not committed to shifting to the 800-volt architecture. CEO Elon Musk has questioned the value of the transition, suggesting it’s not worth the cost of redesign just yet—at least for the company’s smaller vehicles like the Tesla Model Y and Model 3. In the long term, though, he believes that an 800-volt architecture would make sense for a high volume of vehicles.

Also, I just caught this from Erickson during the ER call on AEHR’s silicon photonics business:

While we believe it will likely be several years before we see significant revenue generated from this optical chip-to-chip communications market, this order from our lead silicon photonics customer and their request for an accelerated shipping date is encouraging and provide some data to suggest that this market opportunity could happen sooner.

Sooner than “several years” for “significant revenue” is not something I’m going to base an investment on. Does anyone here think they’ll still be a shareholder of AEHR in several years?

22 Likes

XMFBreakerTinker, others - - I skipped over your comment re Aehr receiving a DARPA contract for memory testing way too quickly. Just checked the SEC.gov archives now and
learned some very interesting things –

1 - back in '97 (26 yrs ago!) Aehr sold $41 million of stock to raise money to support their development of a massively parallel memory test system - - from a 1997 Aehr stock sale prospectus - - -

"Aehr Test develops, manufactures and sells systems which are designed to
reduce the cost of testing DRAMs and other memory devices, and products which
are designed to enable IC manufacturers to perform test and burn-in of bare die.
Leveraging its expertise as a long-time leading provider of burn-in equipment,
with over 2,000 systems installed world-wide, the Company has developed and
introduced two innovative product families, the MTX system and the DiePak
carrier. The MTX is a massively parallel test system capable of processing
thousands of memory devices simultaneously. The MTX massively parallel test system is designed to reduce the cost of
memory testing by performing both test and burn-in on thousands of devices
simultaneously, including DRAMs, SDRAMs and SRAMs. IC manufacturers can optimize
the final test process by transferring many time-consuming tests to the MTX
system and using lower-throughput, higher-cost memory testers to perform only
the high accuracy test functions for which they are most effective. The Company
believes that this “mix and match” strategy can enable IC manufacturers to
reduce the required number of conventional memory testers and, as a result,
substantially reduce capital and operating costs. Siemens has purchased
production quantities of MTX systems from the Company, and other leading
manufacturers have purchased units for evaluation.

  1. DARPA supported their effort - - “In 1994, the Company entered into a cost-sharing agreement with DARPA, a
    U.S. government agency, under which DARPA is providing co-funding for the
    development of wafer-level burn-in and test equipment. The contract provides for
    potential payments by DARPA totaling up to $6.5 million. The Company has
    received $2.4 million through February 28, 1997, and the remaining payments are
    scheduled to be made through January 1999.”

My conclusions / thoughts -

  • Aehr made specialized memory IC test machines way back - very surprising to me, and I expect they still have that expertise now
  • Massively parallel means that they built a machine capable of testing ALL of the die on the wafer SIMULTANEOUSLY (while under burn-in conditions). If this was successful (and apparently it’s still a work in process) the concept should greatly reduce the total cost of back-end processing because 1) test cost per die drops since all die are tested at once (today’s best memory test machines I read to be capable of testing about 16 to 32 packaged devices simultaneously, vs. hundreds to thousands on a wafer) 2) packaging costs drop because only known good die end up in packages. Functionally good AND infant mortality screened.

BUT - memory testing is very complex, testing includes functionality, parametric and pattern testing. Each is complex. But it appears that Aehr has that expertise going back quite a few years. AND the ability to do all this testing while also screening for early life failures adds more complexity.

As many have pointed out, Aehr apparently has yet to convince the memory folk. But, I for one, will be less surprised than I would have been a month ago, if this happens at some point. Testing memory die in wafer form is probably the best way to drive down total test cost if massively parallel testing can be successful.

16 Likes

AEHR has legitimate interests getting into memory. Problem is that AEHR has no DRAM customers and no DRAM players even testing out their equipment - at least for memory. Even Gayn indicates it is probably end of decade or later opportunity. For AEHR. We’re AEHR to announce even a developmental shipment to DRAM manufacturer the shares would rocket I’m sure. But it just is a visionary thing and nothing in the real world to indicate when,’if ever, this may come to fruition. So really not something to invest in on that basis unless you know some Micron manager or the like with a wink :stuck_out_tongue_winking_eye: wink deal going on.

6 Likes

@jkemps

AEHR is a company that’s repeatedly been through a lot of pain.

Business was quite good in the 80’s, topping out at $41.8 million in revenue in FY 1989. However, the company didn’t have much cash and carried bank debt. I’m missing a few years of data, but suffice to say that things were not quite as good in the early 90’s. The company was in default on its line of credit and negotiated a profitability covenant waiver.

Things started to get better in the mid 90’s and when AEHR filed for IPO it had closed out fiscal '97 with $42 million in revenue, with positive net income and operating cash flow. However, it had less than $3 million in cash, and close to $5 million in (bank) debt. That’s not an ideal position to be in when you’re a company with 159 employees and operating expenses running at $10-12 million per year.

So, AEHR did not sell $41 million of stock. It went public and raised $26.8 million, and one use of the proceeds was to get rid of bank debt. They had developed the MTX system, which was a next step in the evolution of their product lines:

(From 2014. Source here.)

If there is one theme here, then it is that AEHR has very much been in the “memory” business and failed to stay in that business. So (IMHO), the extremely speculative (IMHO) bet here is that the wafer-level FOX product line will enable them to once again regain business in “memory”. A product line that has been around for a long time:

(From 2013 Source here.)

(FOX-XP is the newest iteration, replacing FOX-15.) For more than two decades, AEHR has had product lines covering both packaged parts and wafer-level testing. Here’s the split between the two:

(For where there’s enough data available. A pain to reconstruct. Awaiting 10-Q for most recent quarter.) Pair this with a look at revenue separating out ON Semi:

To me, these two taken together does not show any signs of being able to sustain a business around memory and logic. Despite being in business since 1977 - developing several innovative solutions, the revenue from “packaged parts” (ABTS) has dwindled and (disregarding ON/SiC) wafer-level solutions has not been able to fill the void.


I don’t have a position in AEHR and don’t want to sound overly bearish. I posted a lot of background and numbers (charts) a while ago, and intentionally tried to keep it informative and without opinion. With the level of (IMHO) speculation in this thread, though, I feel the need to say that in my view AEHR is high risk as it is and would caution against adding potential growth areas into the investment thesis. Just as an example:

“Industry experts estimate that the total available market for wafer […] testing including the consumable probe cards is in excess of $3 billion per year. With our FOX-1 product we are targeting the market segments for ICs with long test times such as flash memories.”

(From Q2 2008 earnings call. Emphasis mine.)

31 Likes

Turns out the terminology here is really confusing. AEHR itself said this confusing thing:

…inverter that uses 24 silicon carbide devices which each have two silicon carbide die in each device package. These 48 silicon carbide devices can drive…

So, in one sentence they say “24 silicon carbide devices” and the very next sentence they say “48 silicon carbide devices.”

7 Likes

In the end this discussion over the number of SiC chips in an EV is talking about products not yet in existence and won’t be until 2026 or beyond. It therefore, won’t be relevant at present. Further, we have no idea how many SiC chips will be produced each year. By 2030 multiple OEMs are targeting that 50% of their vehicles will be electric. Some, like Mercedes (from memory) are shooting for 100% of their vehicles being electric. Tesla is targeting 20 million vehicles being produced. They are now producing somewhere a bit north of 1 million. Tesla is also coming out with the cybertruck and producing semitrucks that are quite cool. Rivian is building trucks and so on. So is there 24, is their 48 per inverter (I believe it is presently 48) does it really matter. What matters more is how quickly and how many EVs are produced. Almost all next generation EVs (defined as those coming ot market in 2024 and beyond) will be based on SiC. That SiC must be orphan tested and burned in.

The more imperative issue is will AEHR equipment be used to test the vast majority of these chips? Silicon photonics and other use cases are still speculative and are nice future S curves for the future and the narrative, but for now it is SiC. AEHR has $100 million this fiscal year if they meet current guidance. This revenue number is prior to next generation EV production outside of Tesla even come to market in any volume. What happens to AEHR revenues when next generation EVs begin to proliferate beginning in 2024 and beyond? Answer is simple, AEHR revenue will go up.

EVs outside of Tesla are hitting some difficulty. VW decreasing production in Germany. Ford delaying their investment in their next generation battery factory. Tesla continuing to outright destroy competitors even before they come to market by having the best specs while lowering their prices. Tesla already is the low cost producer and yet is selling cars at premiums that most competitors cannot possibly make a profit at, particularly not with all the billions they are laying out.

So for me, and AEHR investment is much more about the proliferation of the EV market outside of Tesla than any of the other interesting, and relevant, and important, but not critical minutia we have been discussing.

25 Likes

Raylight - thanks, very, very enlightening. After your post and data, I have to agree - very speculative to assume Aehr will break into memory at this point. But they have to get an award for - it seems - never giving up :slight_smile: Great post, thx, JK

6 Likes

@Smorgasbord1 at the risk of adding nothing worthwhile to this thread I will only say, yes you are right. The 144 SiC chip count was not entirely made up by me, I had read it elsewhere and elaborated by suggesting it was the result of added inverters. I made the same mistake with the comment I attributed to Musk about maybe reducing SiC at an indefinite future date. Again, read it elsewhere and failed to validate it. The interwebs is full of information, much of it is incorrect. I’ll be more careful in the future.

8 Likes