You won't believe this, but Coby from Weebit has again asked us to reschedule...

For now, I think we'll just leave this. They are proving to be too unreliable.

(for the record, each scheduled event was at a date and time of their choosing)

:( super disappointing from Coby -- understand that he is super busy but multiple postponements is a bad look.

Shame ......... It might have been a great opportunity to see what the heck is behind today's massive pullback!!!

@slymeat The primary intent of my earlier post was to encourage more critical analysis of the third-party information/commentary being shared on SM in support of the Weebit thesis. My critique of Weebit-related commentary was not intended as a personal attack, but was perhaps clumsily worded and not specific enough to be constructive for which I apologise. I'll try do better in this post, and hope that my words are read in the spirit of good faith critique/discussion in which they are intended and not as a personal attack.

In the interest of continuing a more concrete and productive discussion, I'll do a quick pass over the Electronic Specifier article and your straw to provide a more detailed and hopefully useful critique. I'll also note again that I am no expert in this space and have more questions than facts/answers to offer.

--------------------------------------------------------------------------------

Publisher

The article was published in a free magazine put out by ES, who would appear to use the publication as a way to earn revenue from electronics suppliers looking to generate product awareness and sales. The article provides no editorial scrutiny or context at all, and simply lets Weebit's Director of Product Marketing define the narrative, which is a partial regurgitation of a page on their website.

Flash

The article and many Weebit straws talk about flash as if it is one technology with a specific set of properties, but it is in fact an umbrella term for a whole group of technologies that have been continually refined over decades e.g. NAND vs NOR, single- vs multi- vs triple- vs quadruple- level cells, 2D vs 3D spatial orientation, different chemistries and fabrication processes, etc.

To make assertions like flash is an old stagnant technology, or that "typical flash" draws 100x more power during programming, demonstrates a lack of nuanced understanding and/or wilful oversimplification.

NVM use cases

There are probably too many NVM use cases with differing requirements to count. The article claims to be focused on the data logging space, which is barely a focus at all given the breadth of data logging use cases and requirements. For example, I would guess most data logging use cases don't have to worry about high levels of radiation, so this asserted benefit of ReRAM is only narrowly relevant. You assert that the article explains how ReRAM is a superior technology to flash without any qualification of use case and requirements. At best, the article could be said to make a case for considering the use of ReRAM in some data logging applications once it becomes commercially available.

NVM addressing & programming

As far as I understand, your commentary on flash erasure/addressing and suggestion that Weebit's selector somehow offers an advantage to ReRAM over "flash" in this regard conflates a number of concepts and is misguided.

Addressing in this context is a logical function that gives host software an abstract, common way of referencing some finite amount of data (exact number of bits depends on system architecture) in the memory array. The array controller is responsible for mapping an address to the physical cell(s) in the array that hold the data.

Let's baseline our mental model of a NAND flash device using this (simplified) diagram and description which I believe is sufficiently accurate for the job. One relevant detail not shown are the individual cells within each page, where each cell can hold one (SLC) or more (MLC,TLC,QLC,...) bits.

I believe Weebit's selector operates at the cell level i.e. selects one or more individual cells within a page to be active for a given operation. Taking them at their word, their research results suggest that their selector design allows for tighter spacing of cells than traditional transistor-based selector designs, and therefore better physical density of cells within a given area of silicon.

That is indeed a fine thing, but their selector has no bearing on the number of pages that read/write/erase operations apply to, or the number of addresses required to reference a given amount of data. For various technical reasons, I assume ReRAM cells will still be arranged in pages, pages into blocks, etc. similar to typical NAND flash devices, and therefore the erase operation will still not be capable of operating on an individual cell.

Endurance

There seems to be confusing guidance and contradictory statements about cell endurance. Weebit claim 10x-100x better endurance than typical flash's 10k cycles, but the production SkyWater specs state 10k-100k cycles. There's also a big difference between 10k and 100k - which is it? Their endurance superiority claims also appear to be FUD given that a Google search for "high endurance nand flash PE cycle" comes up with plenty of reading and product pointers.

The other thing about endurance is that it has become less and less of an issue as controllers (and the firmware which runs on them) have been improved over time. Intelligent wear leveling ensures that for many write-intensive applications, the endurance of the flash is not the primary factor limiting the device's useful working life.

Security

I would characterise the article's security discussion as FUD. A core tenet of digital security best practice is that if someone can gain physical access to your device, most if not all bets are off i.e. if someone can get close enough to monitor electric field emissions from a comms bus or shine a laser beam on your NVM, you've got bigger problems.

If someone can gain remote access to your device because it is attached to some sort of network, the physical attacks described are presumably not possible and you're reliant on software/information security mechanisms such as process isolation and data authentication/encryption rather than the electrical characteristics of your NVM.

Cost

I assume the cost delta asserted is theoretical given that the well trodden manufacturing processes and economies of scale that exist today for commodity types of flash don't exist today for ReRAM production?

--------------------------------------------------------------------------------

I'll pause here and welcome any further discussion.

Continued some casual spelunking today, this time to look into Weebit's claims about process limitations for flash (infographic taken from here).

First up, it's important to differentiate IC designs which call for embedded vs external NVM.

The former (referred to as "eFlash" in the Weebit infographic) calls for the NVM to be etched onto the same die as other circuitry that will make direct use of the memory e.g. an ASIC design can place a processor, DRAM, NVM and IO devices all on the same die.

The latter refers to the NVM and its associated control logic being etched as a dedicated IC or set of ICs grouped on a PCB, that is then connected to some sort of bus on which other devices can access the storage e.g. NVMe computer storage is a crap load of NAND flash + a controller sitting on the PCI express bus that the host CPU and software can access.

For the external case, NAND flash production would appear to scale down to modern fab processes just fine e.g. Samsung's 10nm process flash.

For the embedded case, Weebit claim eFlash hits a wall below 40nm or 28nm depending on a roll of the ASX announcement dice (I'll assume the discrepancy is an honest mistake presumably from reuse of older slides/text when 40nm was the commercially available lower bound). My spelunking suggests that there may not actually be a wall below 28nm, but there will be increasing difficulty getting eFlash to work in newer processes and there does indeed appear to be a broadening of the NVM tech landscape underway (e.g. ReRAM, MRAM, PCM). Having said all that, I did come across references to eFlash being able to scale well below 28nm e.g. 14nm.

Which brings us to the competitive landscape. Let's put aside the fact that ReRAM is only one technology in the broadening NVM tech landscape, and it's unclear to me if it's a winner takes all/most race or a different horses for different courses type affair. I come across this write up of TSMC's 2020 technology summit in my travels which has some useful commentary and infographics. Of particular note is that TSMC has a 22nm embedded ReRAM offering ready to go and is looking to 16/12nm as the next step.

I don't know anything about Weebit's intellectual property portfolio, but if TSMC has brought 22nm ReRAM to market without having to licence anything from Weebit, does Weebit have any kind of useful moat? And if we put the moat question aside, what does the overall ReRAM market share and Weebit's sub-share end up looking like when all the major fabs (TSMC, Global Foundries, Samsung) appear to be offering embedded Flash, MRAM and ReRAM?

Other random/interesting things I stumbled across during my spelunking:

• A patent application for forming 3D NAND in BEOL processing (Weebit claim that flash is integrated in FEOL which has some drawbacks relative to ReRAM's BEOL integration).
• Automotive use cases appear to be a significant driving force for smaller process scaling and pose significant environmental operating requirements on NVM.
• An interesting slide pack looking at ePCM for automotive use

Good commentary @loshell.

"I don't know anything about Weebit's intellectual property portfolio, but if TSMC has brought 22nm ReRAM to market without having to licence anything from Weebit, does Weebit have any kind of useful moat? And if we put the moat question aside, what does the overall ReRAM market share and Weebit's sub-share end up looking like when all the major fabs (TSMC, Global Foundries, Samsung) appear to be offering embedded Flash, MRAM and ReRAM?"

Yes, it absolutely has a moat -- provided there are no issues with scaling down and competing with TSMC's ReRAM. TSMC already have 40nm and 22nm developed and production ready. One of the main opportunities for Weebit is, due to TSMC being a foundry, the only ones that will benefit from TSMC ReRAM is their customers. Where does that leave the other foundries that are competing with TSMC, who want to integrate ReRAM into their offering? They will require RERAM tech of their own and they obviously can't use TSMC's. Further, what about Intel, NVIDIA, STM etc. -- all will potentially have interest in integrating non-TSMC ReRAM into their products.

Another key factor is WBT's flexibility to work with customers around providing tailor-made solutions. TSMC provide an off-the-shelf product (from what I understand anyway).

This provides an enormous opportunity for Weebit, but execution is key here -- a long way to go. That said, they haven't disappointed in my opinion. I am backing them.

@Rocket6 but it doesn't appear to be an exclusively TSMC vs Weebit ReRAM race e.g. Global Foundries partnering with Dialog Semiconductor to offer a ReRAM option to their fab customers. If Weebit are not the sole, permissively licencable ReRAM IP in town, then the argument that all other fabs which want to offer a ReRAM option to their customers and not invest in creating their own IP will have to licence from Weebit doesn't hold up.

The customisation aspect is definitely something likely to be of value to some potential chip designers and/or fabs, but isn't a mass market commodity production gravy train either. Volume licensing fees for off-the-shelf commodity designs is presumably the hope/expectation for where the rivers of gold will eventually flow from?

(I promise I'm not trying to be a Debbie Downer for the sake of it, and genuinely hope that my increasing suspicion and scepticism are not justified!)

No worries @Macca571 - happy that my explanation was helpful. As expected, WBT was reinstated today, and they went straight back to the north-bound SP trajectory they were on before the recent "issues", finishing the day up +5.49% (or +26 cents) at $5.00/share. The reinstatement announcements came at 2:33pm this afternoon.

From the ASX: WBT-Reinstatement-to-official-quotation-from-ASX.PDF

And from WBT: Weebit-Nano-resumes-trading--'Meet-the-CEO'-details.PDF

The link to pre-register for their "Meet the CEO" virtual interview @ 4pm AEDT on Monday (6-Feb-2023) won't work by clicking on that screenshot above, but you can click on the PDF file link above it and then click on the link within the file, or here - I'll reproduce that link now:

To attend the event, please pre-register at: https://us02web.zoom.us/webinar/register/WN_P3OFN1NFRz-Cu3z8tDgtxQ