AEHR and SiC usage in EVs

Investment Analysis Clubs Saul’s Investing Discussions
1

After some discussion and a thorough reading of the various threads here, I did an internet deep dive on SiC usage in EVs, and where AEHR’s test products fit. Some of this comes from this IET (Institution of Engineering and Technology) reseach paper, which is pretty dense:

Some from a LinkedIn post and comment thread:

Some from this article:

Some from three analyses of Tesla’s claim to reduce SiC usage by 75% in future vehicles:

A GaN vs SiC comparison article here:
https://www.avnet.com/wps/portal/us/resources/article/understand-practical-gan-and-sic-differences-for-ev-onboard-chargers/

SiC Failure Rate analysis from Wolfspeed and others:

And even a little bit from:

OK, references out of the way, here’s my investor-oriented summary:

  • All EVs have inverters, which convert the DC output of a battery pack to an AC output voltage that drives the AC electric motor. The inverter controls the magnitude and frequency of that AC power, which can easily exceed 200kW (200,000 watts).
  • Tesla pioneered the usage of SiC (Silicon Carbide) in its inverters for the Model 3 in 2017. Before then, almost all EVs/hybrids used IGBTs (Insulated Gate Bipolar Transistors) to control their high power traction motors.
  • The advantages for SiC include higher efficiency (5%-15% overall), smaller package size, less produced heat, and longer life (about 80% longer). In particular, higher efficiency means the vehicle achieves a greater range with the same battery, or can reduce costs by having a smaller battery with the same range. Smaller packaging helps with aerodynamics and interior/cargo space.
  • SiC can also be used in vehicle On-Board Chargers. The advantages here are similar: higher efficiency (less of the input power is wasted), smaller packaging size (28%) and weight (24%) in the vehicle, and longer life. However, this application doesn’t have the price-offsetting factor of potentially reducing battery size. That said, in 2019 Tesla changed their On-Board Chargers to use SiC, and that resulted in higher charge rates at Level 3 chargers.
  • SiC can also potentially be used for DC to DC converters. An EV’s main system runs at either 400 or 800 volts today. But, just about everything except the traction motor and AC compressor runs at 12 volts. There are some advantages here, but since this isn’t a big power draw compared to making the vehicle move, efficiency gains are probably not worth the cost.
  • A note on GaN. While inverters can use GaN, the technology isn’t yet there for use in EVs. First is what appears to be an inherent limitation of about 650volts. This is fine for most EVs that use 400 volt systems, but there are already EVs from Porsche, Lucid, Hyundai and McLaren that use 800 volt systems. It seems more likely that GaN usage in EVs will start with the On-Board Charger, but again for automotive cost and availability are very real concerns. I will note that today you can buy GaN “wall warts” for charging your USB devices today. They are smaller/lighter and can put out enough to charge a laptop and other devices simultaneously. I own a few.
  • Earlier this year, Tesla’s VP for Powertrain Engineering, Colin Campbell, made a claim that Tesla will reduce its use of SiC per vehicle by 75% in the future. This sent AEHR and other SiC related stocks down, but they recovered pretty quickly.
  • What I’ve gathered is that Tesla wanted to make sure that their untested SiC-based inverter didn’t fail, and so they used more SiC chips (48 total, with pairs connected in parallel to reduce current through each), and running them well below their maximum temperature range. It’s possible that in later designs (Model 3 as well as the later Model Y not to mention “refresh” versions of Models S and X), Tesla was able to reduce SiC usage based on millions of miles of real-world testing. Separately, teardown expert Munro reported that Tesla has reduced the number of temperature sensors on some of its boards, presumably based on its analysis of temperature data. It’s possible that Tesla has temperature sensors on its inverter module and since Tesla vehicles upload data back to Tesla, they could have done an analysis of what temperatures the inverters run at, even when pushed, and made internal design changes.
  • Tesla used what is know as “Gen 2” SiC chips in its first Model 3 vehicles. We are now at the 3rd generation of SiC MOSFETs, which have higher current/temperature ratings. This alone could count for 50% of the Tesla target of 75% reduction is SiC count.
  • Two other scenarios for SiC count reduction have been discussed. The first is that Tesla’s upcoming mass-market vehicle will have lesser performance characteristics, and thus won’t need to handle as much power, thus reducing the need for parallelizing chips to reduce current/temps. Another is speculation that since Tesla already uses multiple motors/inverters in its products today (both “dual motor” and the top-end Plaid’s “Tri-motor” vehicles), that Tesla would actually revert back to regular Silicon for the motors that kick in on high power demands. The idea is that there’s a main motor that handles everyday driving, and that uses SiC for efficiency. But when the driver puts their foot down, the secondary/tertiary motors that then get engaged don’t need to be that efficient since they’re hardly used. Presumably, a 6% reduction in 50% of the power wouldn’t be missed, and overall range is still preserved by the main motor still using SiC. I don’t know what to say about either of these.
  • Although it’s been 6 years since Tesla first used SiC, you can still buy vehicles today that aren’t using this technology. That’s going to change as these vehicles simply aren’t as competitive in the market.
  • One of the main advantages of AEHR’s testing is that it’s done at the wafer level. This is important, because once the chip is packaged into its device, swapping them out is expensive, if even possible at all. And even if you had a low failure rate, with dozens and dozens of these chips per inverter, the overall rate of failure is multiplied many times over. I don’t know the failure rates, but consider that if the failure rate for SiC chips were 1/10,000, then with 48 chips about 5 of every 1,000 inverters you build will fail. That’s not acceptable, and I believe the failure rate is worse than 1/10,000.

Adding one more bullet point here:

  • A third speculation for Tesla reducing its SiC usage is from adopting a new type of SiC architecture called a “Trench.” Apparently, the existing designs are all “Planar” architecture. The Trench architecture has a higher power density, so fewer chips would be required. One site thought this “unlikely” as it breaks from the current norm, but then Tesla broke norms by using SiC in the first place.

Here’s a tweet on this architecture change ramification:

So, it may be more likely than some SiC insiders think. What this means for AEHR is beyond my knowledge, unfortunately.

65 Likes
2

I’m adding a second post here because one of the sources I used above has some actual dollar numbers on AEHR’s products as of April this year:

Here’s a link to one of the slides from AEHR:

1456×817 105 KB

Note the low cost per chip in terms of equipment depreciation.

In a recent video linked by @BroadwayDan, the CEO reiterated that Apple was one of its 10% or more customers. That appears to be for testing of Apple’s FaceID chips.

The article I linked above says an SiC wafer could have up to 1,000 devices and AEHR’s FOX-XP system can test 9-18 wafers at a time at an initial cost of $2.5m USD.

Next let’s consider the “consumables.” Now that I understand things better, that’s not the word I would use at all. What AEHR calls a consumable is their “WaferPak” connectors. These are basically like “probe cards” for the system. Probe cards are printed circuit boards out of which tiny probes extend that contact the wafer so that things can be measured. To fill out a $2.5m FOX-XP you’ll spend about $1.5m on WaferPaks.

Now, to be clear, WaferPaks don’t appear to wear out. You don’t need to buy new WaferPaks after N usages or anything like that. What happens is that as an SiC chip manufacturer, you improve your chip design every few years. Since the WaferPak has to touch the wafers at exactly the right spots, you’ll need a newly designed WaferPak for each new design.

So, what AEHR really has here is a modular design so that your entire investment in test machinery doesn’t get thrown away when you redesign your chips. Spend $1.5m instead of $4m for a whole new setup. That’s good, but it’s not consumables. And this shows why I don’t think we will see much “consumable” revenue for a few years.

Another component that AEHR sells is the “WaferPak Aligner.” This is used as part of automation to load/unload wafers. This is an additional $1m product (about), and is not a consumable. It’s not required, but it seems any volume user would want these to automate testing and reduce downtime of the testing machine.

Anyway, that’s some additional insight, with dollars, on the testing products and so-called “consumables.”

67 Likes
3

@Smorgasbord1, Thanks for the deep dive. I’ve not read the reference material for the links you provided, but I’ll probably just rely on your summary.

Problem is, I don’t really know what to make of your summary with respect to Aehr’s revenue. The clarification about “consumables” was enlightening. During the 1Q24 conference call Gayne said (or implied - I need to re-read it) that chip redesigns are happening much more frequently than “every few years.” If I remember correctly he said that the chips are redesigned multiple times in a year.

Apparently, SiC automotive chip design is a balancing act. There’s a lot of tradeoffs with respect to power handling, charge time, drivetrain performance, heat dissipation and the list goes on. With real world experience the car makers frequently adjust the balance of these factors which means the wafer changes and a new WaferPak is required.

I don’t know, but I think the car makers will probably settle on one, or a few designs and the demand for WaferPaks will go down. But for now, I think the redesign activity will remain high.

All things considered, I must admit that the investment thesis for Aehr is much more complicated than I previously thought.

15 Likes
4

Does every chip redesign imply a new outside interface, i.e., the need for a new WaferPak?

2 Likes
5

@Tamhas, The simple answer is I’m not sure, but I think “yes” is most likely the correct answer, or maybe much more often than not is the right answer.

The WaferPak must make numerous precise contact with each chip on the wafer. During the 1Q24 CC in his prepared commentary Gayn said:

“We’re also very pleased with the continued stream of new designs for WaferPaks we’re seeing. Our new design volume has tripled over the last nine months as we’re seeing more and more electric vehicles coming online with their own specific device designs for inverters and onboard chargers. As a result, our customers are buying additional WaferPak Contactors for these new designs, highlighting the recurring revenue part of our business.”

The discussion about WaferPak design goes on for another couple of paragraphs. They are very pleased because each time they receive a new chip design, they have to design and make a new WaferPak that conforms to that new design. Gayn further commented that they are adding resources in both the US and Philippines because of the volume of new chip designs. In 1Q24 WaferPak sales represented over 50% of total revenue.

As I understand this a separate WaferPak is required for each wafer, so a FOX-XP configured to process 18 wafers simultaneously requires 18 WaferPaks. From what @Smorgasbord1 wrote in this thread, the WaferPaks required for an 18 wafer FOX-XP is $1.5M, that’s quite a lot of money.

As @Smorgasbord1 pointed out, calling the WaferPak “consumable” is somewhat misleading (unless they wear out with use, which I doubt). It’s not that they are actually consumed over time and must be replaced. What actually happens is that they become obsolete each time the chip design is revised so a new pack of WaferPaks is required to accommodate the new chip design.

4 Likes
6

That’s not my read of it at all. Here’s a quote from the transcript:

During the quarter, we had record shipments of our FOX WaferPak Contactors in both revenue and units with revenues reaching well over 50% of total revenues for the quarter. We’re also very pleased with the continued stream of new designs for
WaferPaks we’re seeing.

Our new design volume has tripled over the last nine months as we’re seeing more and more electric vehicles coming online with their own specific device designs for inverters and onboard chargers. As a result, our customers are buying additional WaferPak Contactors for these new designs, highlighting the recurring revenue part of our business.

and

As our customers win new designs from their customers, Aehr eventually secures orders for new WaferPak to fulfill these new wins.

I saw nothing about frequent redesigns in the transcript, just that different OEMs may be making different demands on their Tier1s for what their inverters and chargers need to be, not that they’re going to change.

Here’s an interesting video snippet to watch. It’s Sandy Munro taking an inverter out of a Model S Plaid and walking over to their Model 3 Front, Model Y Front, and Model 3 Rear front and Model Y rear motors, showing how the same inverter package fits all those motors as well:

Heck, even the boards in the Model S plaid inverter say “Model 3.” Sure, some of the chips are modified, but I suspect the SiC chips are not. If anyone were to frequently redesign their chips, it would be Tesla, who is famous for continuouos improvement.

I agree. I have not sold my position out or even down, but am re-considering what my investment thesis is.

24 Likes
7

My thought was that the WaferPak connected to the contact points on the chip and that considerable change in internal logic could occur without changing the position or function of the contact points.

1 Like
8

Smorg,

I wanted to thank you for all the digging on this stuff. I know there is 2 threads with a lot of good info so if I missed anything forgive me. This is the Press Release from AEHR the day Tesla announced the 75% reduction.

Some bolded points from the PR:

“In summary, we believe the announcement by Tesla does not impact the market significantly, either higher or lower, as they are adding a new half power drive train to be used on a new lower cost platform that will increase the market opportunities for Tesla and help drive volumes particularly in lower cost target markets such as China, but also in the US and around the world. The move to larger die to be package in module form is seen as a benefit for Aehr’s wafer level test and burn-in solution total available market."

The PR goes into details on what Tesla models and inverters the reduction/changes apply to.

“Aehr believes it makes sense and stands to reason that Tesla would make a lower cost lower power inverter and drive unit for a new low-cost platform. An inverter made of 12 100A silicon carbide devices would provide a 200A, 400V, 3 phase 80 kVA inverter and create a lower cost drive unit that would be only half the power of the current drive units in current models. In Aehr’s opinion, putting a Model 3 inverter into a lower performance “Model 2” vehicle seems like overkill and unnecessary. It is also important to understand that a 100A device using today’s generation of silicon carbide devices is approximately 50% larger than the devices used in the current Telsa inverters. So, while this new lower performance 800 kVa inverter only uses 12 die or 75% less than the current 48 die design, the die themselves are estimated to be about 50% larger, or require 50% more wafers for the same number of die.”

“In addition, during the Q&A session, Tesla further clarified that the new inverters would be made from a new Tesla-proprietary custom module package, and that Tesla would purchase the die from multiple manufacturers and package them in this Tesla-proprietary custom module. Again, Aehr sees this as a natural roadmap and consistent with the roadmaps stated by major manufacturers of silicon carbide where the electric vehicle inverters will migrate multi-chip modules to reduce power conversion losses, improve thermal performance, simplify design, and lower overall cost of the inverter system"

It seems like AEHR management is well aware of what Tesla is doing & not in denial. They are expecting these reductions from EV manufacturers as part of the natural SiC lifecycle. The stock was hit 35%+ the day Tesla announced this and the stock recovered nicely. They reiterated their FY guidance at the time and gave guidance for this FY knowing this info.

Bnh

20 Likes
9

It is not just Tesla driving the growth in the SiC EV market. Just today there was an announcement that Mitsubishi and Denso Corp. will be investing $1B Coherant’s SiC wafer business. This is likely why AEHR is up 14% today so far.

Bnh

16 Likes
10

I watched the whole Investor Day presentation previously, and just now rewatched the entire Q&A session, and heard nothing on SiC/Inverters. Maybe AEHR is talking about some other presentation? It does seem logical that Tesla would design its own module/package, however, and might even manufacture the package.

I think it’s also important to listen carefully what Colin Campbell of Tesla said in his presentation. He didn’t, for instance, say “fewer SiC chips,” he said “we figured out a way to use 75% less.” Is he talking chip count, or is he talking SiC material? I admit it’s hard to know for sure.

What AEHR is saying about size is logical only if there have been no improvements in SiC technology since 2017. Tesla runs 50amp “devices” today, and AEHR is speculating Tesla will run 100amp “devices” in the next gen. But, then says that a 100amp “die” will be twice as big as the old 50amp “die.” That implies no improvements in SiC capacity, that size is the determining factor and whether they’re using larger chips to handle more the same raw material is needed. Given that there’s a new “Trench” architecture coming, I wonder if AEHR is assuming Tesla’s new chips are still the “Planar” architecture. The tech is way over my head, but it does seem the new architecture is more efficient, so perhaps there is some reduction in SiC content achieved.

In addition, Campbell demonstrated that upgrades to the SIC-based chargers in Tesla vehicles that included heat-sink packaging reduced the SiC usage in them. Perhaps that tech hasn’t yet come to the inverters, and that’s also part of the 75% reduction Tesla says they’ll achieve.

I’ll not argue that the number of vehicles is growing faster than reduction in SiC content per vehicle. That does mean as we try to predict growth in SiC testing we need to be careful to not just use growth in number of EVs as a proxy for SiC growth. SiC growth will be less than EV growth, but it is still growth.

The question for me today is since both AEHR and two analysts have about the same $80m revenue target for the next 3 quarters, do I feel that somehow growth will be faster/higher than that? And if so, why?

9 Likes
11

If you compare last year they guided for $60-$70 million and came in at $74 million. They reiterated this guidance mostly last year and then exceeded it by a bit. This year there is more chance for upside as next generation EVs start to roll out. There is also a chance that AEHR raises guidance to say $100 to $110 million as they get more visibility. But I would not count on more than maybe a 5 to 10% beat. This changes if there is a sudden acceleration in demand as OEMs begin rolling out their next generation EVs en masse. But absent this AEHR does not low ball very skillfully. They tend to be straight shooters in regard but still hoping to over-deliver.

2 Likes
12

OK, so then what’s the investment thesis? If AEHR is priced for $100m revenue, and $20.6m was achieved first quarter, and maybe they’ll get to $110m for the year, how is this going to be a market-beating investment for the next 6-9 months?

Is it simply that there will be more SiC usage or sooner than even AEHR thinks is coming? Erickson says the sales cycles are long, so given his long experience in this field, it would seem he’d have a good handle on what will close in the next 2 or 3 quarters.

We already know what Tesla is going to do, and they’ve been cutting prices to meet their P&D goals. So, is Mercedes or Ford or Lucid or Rivian or somebody suddenly going to start selling more EVs such that they beat up on their suppliers to make more SiC chips, and those suppliers will immediately (= 6 months in this world) turn around and buy more FOX-XPs than they were planning?

9 Likes
13

I am well versed in the electric vehicle market. Most EVs sold today other than the new premium models from new OEMs and from Hyundai and Porsche are selling without SiC. It is the new models that are planned to start rolling out in 2024 and accelerating into 2025 and 2026 that the existing legacy OEMs are planning to roll out with goals of their production becoming 50 to 100% of their total production of cars by 2030. An amazing growth rate and accelerated by government subsidies and government regulations. So the expected market (should EVs come out as planned and sell) will be an enormous increase in demand for SiC chips.

SiC chips are in short supply. OEMs are making long-term contracts with SiC manufacturers like ON or Wolfspeed. GM, Mercedes, and Land Rover as an example have long-term contracts with Wolfspeed. Renesas entered into a 10 year SiC supply agreement with Wolfspeed that includes a $2 billion upfront deposit. The SiC market is building out like mad fabs around the world. Wolfspeed alone is investing $6 billion in new fabs in North Carolina, New York, and in Germany.

Point being, either the legacy auto OEMs who are spending $10s of billions of dollars to invest in EVs and fight for supply, and the semiconductor companies who are investing billions in new fabs and buying up old fabs to convert to SiC production with billions more are completely wrong, resulting in one of the worst investing mistakes by an industry in U.S. history (which is possible, particulary since so much of this is government mandated or subsidized) or SiC consumption just for EVs alone (not to mentioned other uses) is going to skyrocket from current numbers.

The reason you invest in AEHR is due to this fact. Real money and plans are in place. The consumer may reject these products but to “save the planet” and not miss out, this investment and these contracts are in place. They are telegraphed. 2024 is when it starts to really roll out and 2025/2026 it becomes mainstream.

As such, OEMS are not able to “beat up” their suppliers this time. SiC supply is predicted to be chased as demand out does supply through the end of the decade. This is a whole different investing topic, but also applicable to AEHR.

Thus, the reason you invest in AEHR is because every single one of the SiC chips that go into an EV has to be burned in and orphan tested. EVERY ONE. Will this be done on AEHR machines? If so, you can compare current demand for AEHR testing of SiC for EVs vs. what will happen in the telegraphed future unless the EV industry, other than Tesla, falls flat on their face.

33 Likes
14

Unless you/we think that AEHR’s Erickson and WS are missing that market demand will produce greater than $80m for AEHR in the next 9 months, why should I invest in AEHR today? It seems an investment today would ideally be predicated on upcoming announcements of AEHR deals that aren’t yet factored into guidance. In other words, that well-experienced Erickson doesn’t understand his sales pipeline right now.

Just to be clear, I agree that the demand for SiC will grow. One question though is whether anyone besides Tesla will make enough EVs in the next couple years to matter. Ford sold a “record” 5,872 Mach-e’s last month. Tesla sold more than 10X as many Model Ys alone. For auto companies to scale up their EV production means sourcing batteries as well as motors as well as SiC based inverters (and maybe chargers, if they go that far), and to have demand for those additional vehicles.

Is your AEHR investment something you’re looking at for a 2-4 year return timeframe?

8 Likes
15

I woud say that any SiC based investment today is based on the 2024-2026 time frame. We all know how all those expectations can be built into the price sooner than later. With AEHR it is also possible that the photonics S curve can begin to establish itself as well during that time frame.

So the investment is based upon higher numbers than just this year. As an example, next year’s projection (educated guess) by the two analysts is $150-$160 million. By Q4 of this fiscal year AEHR will give guidance for the following fiscal year, and its guidance should at least be $150-$160 million, and so on. So even if this year’s guidance is baked in, there is still uncertainty on next year. And that is where the upside will again come in and may surprise higher as AEHR’s customers have to stop trialing and make actual purchase commitments if they want to deliver SiC to their OEM customers.

And yes, I agree, it is not certain that legacy OEMs can compete at all in EVs. This said, they are going to try like hell to do so with their next generation cars (most of them are still selling the Bolt as an example, that has no SiC in it) and someone needs to supply this SiC. And those suppliers will need equipment like AEHR to deliver.

9 Likes
16

AEHR will have something like 100m in revenue and ~30 (maybe 35) million in profits this year. If that was the case each year, they wouldn’t be worth more than half a billion. As they’re trading at 1.2b or so, more is expected.

Will “more” be 150-160m next year as analysts expect? Who knows? And will that be enough? No…more will be expected the next year.

Right now, the market isn’t sure whether to call BS on the next year number, and is even less sure whether or not to believe that they will deliver even more revenue and profits the following year.

It’s a nice story, but there’s a tremendous amount of uncertainty. I’m at a ~6% position and I’m questioning whether or not that’s too high.

Bear

43 Likes