4DS MEMORY LIMITED — Investor Presentation 2024

Mar 3, 2024

4DS Memory Limited | ABN 43 145 590 110 Level 2, 50 Kings Park Road, West Perth WA 6005 PO Box 271, West Perth WA 6872

+61 8 6377 8043 | david@4dsmemory.com | www.4dsmemory.com

==> picture [52 x 37] intentionally omitted <==

ASX RELEASE

4 March 2024

4DS Australian Open Briefings Presentation

4DS Memory Limited (ASX:4DS) (4DS) (the Company), is pleased to attach the Investor Briefing Presentation for the Australian Open Briefing to be held this week in Sydney, Melbourne and Perth.

The event host will be 4DS Executive Chairman Mr David McAuliffe, who will be joined by 4DS Chief Technology Officer Mr Ting Yen and Strategic Advisor to the Board Mr Peter Himes, who are both visiting Australia from the USA to update guests on 4DS Memory’s progress to date and plans for 2024.

4DS Interactive Investor Hub

If you have any questions on this announcement or any past 4DS announcements, check out our Investor Hub. Like, comment, ask a question and view video summaries on important announcements. Shareholders can link through to the Investor Hub via: announcements.4dsmemory.com

Authorised for release by David McAuliffe, Executive Chairman.

ENDS

Contact information

Investors: David McAuliffe 4DS Memory +61 408 994 313 david@4dsmemory.com

About 4DS

4DS Memory Limited (ASX: 4DS), with facilities located in Silicon Valley, is a semiconductor development company of persistent memory technology, pioneering Interface Switching ReRAM for next generation gigabyte storage in mobile and cloud. Established in 2007, 4DS owns a patented IP portfolio, comprising 34 USA patents granted which have been developed in-house to create high-density Storage Class Memory. 4DS has a joint development agreement with Western Digital subsidiary HGST, a global storage leader, which accelerates the evolution of 4DS’ technology. 4DS also collaborates with imec, a world-leading research and innovation hub in nanoelectronics and digital technologies. The combination of imec’s widely acclaimed leadership in microchip technology and profound software and information and communication technology expertise makes them unique.

For more information, please visit www.4dsmemory.com

Disclaimer

This release contains certain forward-looking statements that are based on the Company’s management’s beliefs, assumptions, and expectations and on information currently available to management. Such forward looking statements involve known and unknown risks, uncertainties, and other factors which may cause the actual results or performance of 4DS to be materially different from the results or performance expressed or implied by such forward looking statements. Such forward looking statements are based on numerous assumptions regarding the Company’s present and future business strategies and the political and economic environment in which 4DS will operate in the future, which are subject to change without notice. Past performance is not necessarily a guide to future performance and no representation or warranty is made as to the likelihood of achievement or reasonableness of any forward-looking statements or other forecast. To the full extent permitted by law, 4DS and its directors, officers, employees, advisers, agents and intermediaries disclaim any obligation or undertaking to release any updates or revisions to information to reflect any change in any of the information contained in this release (including, but not limited to, any assumptions or expectations set out in the release).

You should not place undue reliance on these forward- looking statements. Except as required by law or regulation (including the ASX Listing Rules) we undertake no obligation to update these forward- looking statements.

Page 1 of 1

==> picture [139 x 55] intentionally omitted <==

MEMORY FOR THE AI AGE THE MEMORY THAT BREATHES

MARCH 2024

Artificial Intelligence

Nvidia Co-founder and CEO Jensen Huang described the emergence of artificial intelligence (AI) as the beginning of a new industrial revolution, emphasising that there’s an ongoing global ‘awakening’ concerning AI technology.[1]

He also believes that, over the next four to five years, a trillion dollars’ worth of data center infrastructure and hardware will be built across the world.[2]

TSMC founder Morris Chang recently revealed that customers have approached the company to build up to ten new fabs for AI processors, indicating a significant increase in demand for processors used in AI applications.[3]

Apple CEO Tim Cook, at the 2024 annual shareholder meeting, said his company is “investing significantly” in artificial intelligence.[4]

Sources

1. https://www.businesstoday.in/technology/top-story/story/beginning-of-new-industrial-revolution-nvidias-jensen-huang-says-theres-an-ai-awakening-in-every-country-417256-2024-02-13

2. https://www.datacenterdynamics.com/en/news/nvidia-ceo-jensen-huang-predicts-data-center-spend-will-double-to-2-trillion/#:~:text=Nvidia%20CEO%20Jensen%20Huang%20believes,be%20built%20across%20the%20world 3. https://abachy.com/news/tsmc-founder-reveals-unprecedented-demand-10-new-fabs-produce-ai-chips

4. https://www.cnbc.com/2024/02/28/apple-annual-shareholder-meeting-2024-ai-investments.html

==> picture [98 x 41] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

4DS: The Right Memory for the Right Market at the Right Time
----- End of picture text -----

The New Age of AI Processing is built on decades of research into neuromorphic processing and Neural Net modeling These models require huge amounts of data and are the result of billions of weighted matrix calculations

There is an explosion of new processor architectures in development to overcome the speed, energy and memory bottleneck challenges The introduction of a new non-volatile memory with high speed and high density ‘would initiate a revolution in computer architecture’[*] 4DS is that memory, at the right time, for the most important market in the world

==> picture [5 x 5] intentionally omitted <==

==> picture [52 x 61] intentionally omitted <==

==> picture [94 x 53] intentionally omitted <==

==> picture [5 x 6] intentionally omitted <==

==> picture [10 x 10] intentionally omitted <==

==> picture [7 x 7] intentionally omitted <==

==> picture [5 x 5] intentionally omitted <==

==> picture [6 x 5] intentionally omitted <==

==> picture [43 x 48] intentionally omitted <==

==> picture [9 x 8] intentionally omitted <==

==> picture [35 x 10] intentionally omitted <==

==> picture [69 x 19] intentionally omitted <==

==> picture [5 x 5] intentionally omitted <==

==> picture [6 x 6] intentionally omitted <==

==> picture [10 x 10] intentionally omitted <==

==> picture [7 x 7] intentionally omitted <==

==> picture [5 x 5] intentionally omitted <==

==> picture [127 x 27] intentionally omitted <==

==> picture [99 x 44] intentionally omitted <==

==> picture [36 x 177] intentionally omitted <==

==> picture [60 x 52] intentionally omitted <==

==> picture [5 x 6] intentionally omitted <==

==> picture [35 x 19] intentionally omitted <==

==> picture [6 x 7] intentionally omitted <==

==> picture [7 x 6] intentionally omitted <==

==> picture [7 x 6] intentionally omitted <==

==> picture [7 x 7] intentionally omitted <==

==> picture [9 x 9] intentionally omitted <==

==> picture [9 x 9] intentionally omitted <==

==> picture [9 x 9] intentionally omitted <==

==> picture [7 x 7] intentionally omitted <==

==> picture [5 x 6] intentionally omitted <==

==> picture [7 x 7] intentionally omitted <==

==> picture [5 x 6] intentionally omitted <==

==> picture [98 x 41] intentionally omitted <==

*Source: Institute of Electrical and Electronics Engineers.

==> picture [297 x 540] intentionally omitted <==

----- Start of picture text -----

Generative AI
Transforming the
Compute Industry
----- End of picture text -----

AI refers to a long history of machine learning development that has evolved over the past 40 years in the industry

Generative AI refers to the subset of artificial intelligence techniques that enable machines to generate new content, imitate human creativity, and produce realistic outputs

Machine learning is in everything from your home thermostat to your car’s antilock braking to your Alexa connected device

AI models are built on different types of neural nets which use very large sets of data to create predictive models for different tasks (text generation, image recognition or generation, code writing)

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

The 4DS Advantages
----- End of picture text -----

==> picture [212 x 421] intentionally omitted <==

----- Start of picture text -----

Persistent
Memory at
DRAM Speeds
4.7ns write for 1 hour
read – no refresh
needed
Single Shot
programming
----- End of picture text -----

Tunable Retention from Seconds to Days

Highest energy optimization for warm data applications Can trade off long retention for reduced endurance

Low Energy per Bit

For Warm Data and Persistent Memory Applications

Scalable to Any Process Node

Simple fab processing, compatible with standard tools Low-cost BEOL Integration, compatible to any advanced CMOS logic process

Analog Programmability

By time or voltage modulation

==> picture [98 x 41] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

An Overview of Memory Terminology
----- End of picture text -----

SRAM

Volatile, provides the fastest read time, used for immediate access by CPU. But also large in size and power (6 transistors needed per cell)

DRAM

Volatile, provides large data storage capabilities with fast read and write times. Architected for large data transfer at high speeds. Needs constant refresh to retain data

4DS ReRAM NAND Persistent, provides hours of Non-volatile, provides lowest retention with very fast write cost per byte of storage. time. Operates at the same Architected for very large data speeds as DRAM transfer and storage. Slow write compared to DRAM

MRAM, PCRAM, Filamentary ReRAM

Warm vs Hot Data

CXL or Compute Express Link

Non-volatile, provides high density alternatives to Flash for advanced node processes. Speed and energy vary but generally slower than DRAM

Different classes of data needed on demand by the CPU or GPU

De facto communications protocol to manage Storage Class Memory

==> picture [98 x 41] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

What is PCMO and why is it unique?
----- End of picture text -----

Current is spread across entire interface area Low current densities lead to high cell endurance Top electrode Thin oxide layer FAST Super fast programming Perovskite layer comprised of: One Shot write P E Field injects or raseodymium Calcium removes oxygen from M Area based current anganese PCMO vacancies O xygen High endurance material Bottom electrode

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

continued
What is PCMO and why is it unique?
----- End of picture text -----

High O_ affinity by top electrode creates an oxygen depletion layer at the surface junction (oxygen vacancies in the PCMO lattice)

Pulse of electric field pushes oxygen back into PCMO, filling vacancies (very fast response: EPIR)

When the oxygen is present the cell conducts and is said to be SET This is a reversible process, creating the on/off resistance states of the memory cell

Total current is a function of this oxygen distribution across the oxygen vacancies as determined by the one-shot programming

==> picture [98 x 41] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

ReRAM: Two Different Technologies,
Two Different Memory Solutions
----- End of picture text -----

4DS PCMO

OTHERS – Filamentary

Area based, lower current density programming, high endurance

Filament based, high current density, low endurance for reliable cell

Up to 10[9] Endurance

Endurance can be as low as 10[4]

Analog characteristics

Primarily designed for digital use

Persistent – tunable retention from seconds to days

Long retention and short retention can be integrated on same chip

Non-Volatile – but requires strong error correction or high energy to create multiple filaments Iterative programming needed for reliable cell

Extremely fast EPIR one-shot response to programming signal – 4.7ns write, well within DRAM window

Used as embedded NOR replacement or CXL-based storage memory

Ideal for in-memory compute or Warm Data storage

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

AI is Driven by Deep Neural Nets
----- End of picture text -----

==> picture [876 x 330] intentionally omitted <==

----- Start of picture text -----

Pre-train
Prompt Decode
Please write a Instruction Generative Result Once upon a time,
story about a cat. I AI Models R there was a cat
1 1 named Jesse...
Dog
----- End of picture text -----

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Neural Nets are an approach to modeling
how the brain builds recognition and recall
----- End of picture text -----

They are a Digital approximation of the Analog process that goes on in the brain

==> picture [654 x 319] intentionally omitted <==

----- Start of picture text -----

Action Post-synaptic neuron Action
potential potential
Axon Axon
Dendrites
Dendrites
Pre-synaptic neuron
Synapse
----- End of picture text -----

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

The energy cost of AI: 70-80% of the AI operations
are due to the weighted calculations
----- End of picture text -----

Moving the data in and out of the GPU takes 5x the energy of the operation itself, and bottlenecks the speed of the GPU

==> picture [614 x 269] intentionally omitted <==

Source: B.Fleischer, VLSI’18.

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

The energy cost of AI: 70-80% of the AI operations
continued
are due to the weighted calculations
----- End of picture text -----

The global demand for electricity may see a huge surge because of AI processing

Low case Base case High case

Global electricity demand from data centres, AI, and cryptocrurencies, 2019-2026

==> picture [661 x 250] intentionally omitted <==

Notes: Includes traditional data centres, dedicated AI data centres, and cryptocrurency consumption; excludes demand from data transmission networks. The base case scenario has been used in the overall forecast in this report. Low and high case scenarios reflect the uncertainties in the pace of deployment and efficiency gains amind future technological developments. Sources: Joule (2023). de Vries. The growing energy footprint of AI: CCRI indices (carbon-ratings.com); The Guardian, Use of AI to reduce data centre energy use; Motors in data centres; The Royal Society, The future of computing beyond Moore’s Law; Ireland Central Statistics Office, Data Centres electricity consumption 2022; and Danish Energy Agency, Denmark’s energy and climate outlook 2018.

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

A Major Driver for new AI architectures:
Moving Memory Close to Central Processing Unit (CPU)
----- End of picture text -----

In-Memory Computing is the defining feature of all emerging AI engines

Conventional Near-memory In-memory computing system computing computing

==> picture [637 x 142] intentionally omitted <==

Sources: Mutlu et al ., “A modern primer on processing in memory”, (2022) .

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

SRAM has been the approach for In-Memory Compute,
but Running out of Steam
----- End of picture text -----

Nvidia’s A100 processor has only 40MB of on-chip memory, an order of magnitude lower than what is required

AI and Memory Wall

==> picture [630 x 319] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Which is driving a golden age of emerging processor
architectures and Integrated Circuit (IC) growth
----- End of picture text -----

==> picture [188 x 156] intentionally omitted <==

----- Start of picture text -----

Global Semiconductor
Market Growth and
Forecast
Global Semiconductor
Market Size (USD Bn)
Forecast
----- End of picture text -----

==> picture [643 x 337] intentionally omitted <==

Source: World Semiconductor Trade Statistics and Mckinsey’s projected yearly growth. Past performance is no guarantee of future results. Not intended as a recommendation to buy or sell any securities mentioned herein, or any call to action. For illustrative purposes only. Actual future semiconductor market growth is unknown.

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Which is driving a golden age of emerging
continued
processor architectures and IC growth
----- End of picture text -----

Artificial intelligence (AI) chip market revenue from 2022 to 2027

Nvidia market capitalisation

==> picture [389 x 266] intentionally omitted <==

==> picture [347 x 278] intentionally omitted <==

Source: S&P Capital IQ. By The New York Times

Source: ©️ Statistica, 2024

A need for a New Memory Solution Exists

“Difficult challenges gating development of beyond-CMOS devices include those related to memory technologies, information processing or logic devices, and heterogeneous integration of multi-functional components, a.k.a. More-than-Moore (MtM) or functional diversification.

One challenge is the need of a new memory technology that combines the best features of current memories in a fabrication technology compatible with CMOS process flow and that can be scaled beyond the present limits of SRAM and FLASH. This would provide a memory device fabrication technology required for both stand-alone and embedded memory applications. The ability of a chip to execute programs is limited by interaction between the processor and the memory, and scaling does not automatically solve this problem. The current evolutionary solution is to increase cache memory, thereby increasing the floor space that SRAM occupies on a chip. However, this trend eventually leads to a decrease of the net information throughput. Volatility of semiconductor memory requires external long-term storage media that tend to be slow to access (e.g., magnetic hard drives, optical CD, etc.). Therefore, development of electrically accessible non-volatile memory with high speed and high density would initiate a revolution in computer architecture. This development would provide a significant increase in information throughput ” beyond the traditional benefits of scaling.

Source: IEEE, The International Roadmap for Devices and Systems: 2022, “Beyond CMOS”, chapter 1.2.2. Emphasis added.

==> picture [98 x 41] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

The Speed Advantages of 4DS Memory
----- End of picture text -----

4DS has demonstrated reliable, robust ReRAM programming with a single 4.7ns write pulse

This improves on our previously reported 9.5ns

The sub 5 nanosecond speeds are due to Electric Pulse Induced Resistance switching, which enables single shot programming in the 4DS cell

==> picture [526 x 220] intentionally omitted <==

This switching is faster than the DRAM write window of 30ns and directly translates to lower energy per bit writing for the 4DS cell

Disclaimer: these graphs are based on best publicly available data and may not reflect actual state of the art for each technology. The graph for the 4DS cell is estimated for the 20nm cell, based on scaling from 60nm actuals.

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Persistence + Speed = Energy Savings
----- End of picture text -----

For permanent offline storage, traditional NVM offers the best solution For long persistence data 4DS offers a low energy alternative

DRAM needs constant refresh

4DS needs no refresh within its persistence window, and then can be ‘refreshed’ within the DRAM window (‘hidden refresh’)

==> picture [512 x 305] intentionally omitted <==

Disclaimer: these graphs are based on best publicly available data and may not reflect actual state of the art for each technology. The graph for the 4DS cell is estimated for the 20nm cell, based on scaling from 60nm actuals.

Reference for this data: K, Prall, “Benchmarking and Metrics for Emerging Memory,” 2017 IEEE International Memory Workshop (IMW), Monterey, CA, USA, 2017

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Why Important? Warm Data Applications
----- End of picture text -----

Advanced computing always needs DRAM for constant read and write (Hot Data), but emerging Inference Engines for AI need data at DRAM speeds that doesn’t change over time (Warm Data)

Storage Class Memory – data storage that is peripheral to the main processing data but not in bulk storage – has consolidated behind CXL or Compute Express Link

This supports many classes of memory behind a standardized communications protocol

CXL is considered ‘Slow’ data when compared to DRAM

==> picture [485 x 273] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Why Important? Analog Compute in Memory
----- End of picture text -----

Using SRAM is the common way of tackling the bandwidth problem in AI processing

SRAM is the fastest memory for such uses

However the number of weights is limited by SRAM cell size (6T) and # bits per weight needed A scalable analog alternative could increase total number of weights and simplify the vector operations

Conversion back to the digital domain will still be needed

==> picture [444 x 270] intentionally omitted <==

Source: Chih et al., ISSCC (2021)

==> picture [961 x 78] intentionally omitted <==

----- Start of picture text -----

4DS welcomes Advanced AI processor companies to explore these ideas with us
----- End of picture text -----

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Corporate Summary
----- End of picture text -----

ASX Symbol: 4DS

4DS Memory 12 month share price graph

Shares on Issue 1,762,834,918 Options on Issue 47,207,184 Cash on hand ~ $10M Share price (1 March 2024) 0.110 Market Cap (at 0.110 share price) $193M

Top 5 shareholders own 10.39% (at 1 March 2024)

Citicorp Nominees 3.36% James Dorrian 2.57% KZ 3 Pty Ltd 1.62% Mr John Clement Cowie Love 1.52% Mr Sam Huu Hai Nguyen 1.32%

==> picture [490 x 195] intentionally omitted <==

==> picture [960 x 120] intentionally omitted <==

----- Start of picture text -----

Board and Management
----- End of picture text -----

==> picture [107 x 107] intentionally omitted <==

==> picture [107 x 107] intentionally omitted <==

==> picture [100 x 107] intentionally omitted <==

==> picture [107 x 107] intentionally omitted <==

==> picture [107 x 107] intentionally omitted <==

David McAuliffe Executive Chairman

Experienced company director

Involved in numerous capital raisings and in-licensing of technologies

Founder of several companies in Australia, France and the UK, many of which are now ASX listed. Non-Executive Chairman of Invex Therapeutics Ltd

Dr Guido Arnout Non-Executive Director

30+ years in commercialising electronics technology

Successes include Power-Escape, CoWare, CrossCheck Technology and Silvar-Liso

Howard Digby Non-Executive Director

Former senior roles at IBM, Adobem Gartner and the Economist Group

Director of Cirralto Ltd, Elsight Ltd and Singular Health Ltd

Advisor to a number of technology companies

Ting Yen Chief Technical Officer

30 years experience in commercialising memory technologies

Various roles at Integrated Memory, Netlogic, Integrated Device, Cypress, Paradigm and Philips Research

Peter Himes

Strategic Advisor to the Board

Experienced senior executive in high grown technology firms

Strong focus on innovation systems and strategic alliances

4DS: THE RIGHT MEMORY FOR THE RIGHT MARKET AT THE RIGHT TIME

==> picture [98 x 41] intentionally omitted <==

==> picture [139 x 55] intentionally omitted <==

FOR MORE INFORMATION

Registered Office Level 2, 50 Kings Park Road West Perth WA 6005

Postal Address PO Box 271 West Perth WA 6872

USA Office 3155 Skyway Court Fremont CA 94539

Phone +61 6377 8043 Email info@4dsmemory.com Web 4DSmemory.com