Firstly, I want to say thank you to the 4DS team. A presentation like this has been severely lacking and well overdue.
Secondly, thank you for providing the information that share holders have been asking for.
So lets clear some things up and summarise
1) Retention - clearly not commercial or price sensitive having been released in a non-price sensitive ASX announcement. Everytime this excuse was given it made me cringe as these are basic metrics other semiconductor companies put out openly in all their technical papers.
2) SCM is no longer being considered. New target market appears to be embedded memory, MCU's (vehicles with all their sensors, AI, etc), SoC's (ie. TV's, phones), AI Processsors, IOT, potentially other niche auxillary roles (ie. pre-cache SSD's as a cheaper, more energy efficient alternative to DRAM) etc. So I'll have to politely disagree with Mjs on it being a universal memory, that I actually personally believed it to be.
3) WD and most other traditional memory companies can no longer be considered potential buyers, narrowing the field significantly to NVidia, Intel, AMD, TSMC, ASML, Samsung, Qualcolmm, Broadcom, etc then Chinese companies like Alibaba. Google has just ditched Broadcom to reduce costs but they are after a chip manufacturer. So as embedded memory chip - it somewhat limits who would be interested with the capital we are all hoping for. Could probably reduce it to counting them on a single hand but they are all significant potentials.
4) Being used as analogue memory is still on the table.
5) The value is in the energy efficiency (my opinion).
For example - NVidia has their H100 and A100 chips, manufactured by TSMC. The H100 GPU SXM accelerator card with TSMC's 4nm process consumes up to 700W, similar to a end-users 4090 which can also be used for AI / crypto / etc however with significantly less on-board memory and only 16MB of L1 cache and 72MB of L2 cache. If that could be increased to say 1GB, that truly is significant.
Combine that with these exponential processing requirements estimated below, and the world will either need to bring some fusion reactors online or the hardware needs to become more energy efficient.
Actually leads me to believe this chart that 4DS used (from an energy report - not computing report
https://iea.blob.core.windows.net/assets/ddd078a8-422b-44a9-a668-52355f24133b/Electricity2024-Analysisandforecastto2026.pdf) is not showing the full picture and is a VERY conservative estimate not understanding the future power requirements for the processing required above. This makes the technology 4DS has more impressive.
For example, from OpenAI directly you can see the computer power increasing significantly - you cannot increase that much each month without also increasing energy consumption to power it.
This is total energy consumption of data centers, which includes cloud and AI processing is from 2022 and seems more plausible and inline with power requirements driven by AI. Note that 4DS's estimated chart of low, base and high cases is a very conservative amount in line with "best case" below.
In conclusion, I think we're all better placed with now with where things are at and where things are heading in regardS to commercialisation.
I expect a PL 7 to follow to further demonstrate the consistency in the manufacturing of the chips, proven yield, contiunued scalability, etc.
Another 12 months, and a further 6-12 months on top of that while discussions and negotiations take place with interested parties.
Just my thoughts on the presentation.
(20min delay)
