AI:You're absolutely right — shrinking the geometry from 60nm to...

  1. 1,498 Posts.
    lightbulb Created with Sketch. 1209
    AI:

    You're absolutely right — shrinking the geometry from 60nm to 20nm can often improve certain memory performance metrics, though it also introduces new risks. Based on known semiconductor scaling principles, 4DS’s architecture, and statements from their own documents, here’s a reasoned prediction of what might improve with Platform Lot 6 at 20nm, assuming the process optimizations hold.


    ---

    1. Write Speed – Likely Improvement

    Why: Smaller cells have less capacitance and shorter ion migration paths.

    Prediction: Write speeds could improve from the already fast 4.7 ns (Platform Lot 4) to potentially <4 ns at 20nm.

    Supporting Insight: In Platform Lot 4, write was achieved via one-shot programming using EPIR (Electric Pulse Induced Resistance switching). This interface-driven process benefits from smaller areas as the pulse has less physical distance to influence.



    ---

    2. Power Consumption – Likely Improvement

    Why: Smaller geometry means lower switching energy per bit.

    Prediction: Energy per write/read could decrease by 30–50%.

    Supporting Insight: 4DS’s ReRAM already demonstrated low energy one-pulse switching. With reduced cell area and lower programming voltages at 20nm, power efficiency should improve, which is critical for AI and edge devices.



    ---

    3. Endurance – Possibly Improved

    Why: Lower energy and more uniform switching could reduce material fatigue.

    Prediction: Could increase from ~3 billion cycles (Platform Lot 4) to 5–10 billion if uniformity and control are better.

    Caution: If scaling introduces variability or edge effects (e.g., poor etch definition), endurance could plateau or even decline.



    ---

    4. Retention – Possibly Reduced or Tunable

    Why: Smaller cells can have higher leakage, especially in analog/multi-level retention.

    Prediction: Shorter retention at default, but tunable by programming voltage/time as demonstrated by 4DS.

    4DS Feature: They explicitly stated retention is tuneable, which suggests they can optimize for retention vs. speed depending on use case.



    ---

    5. Analog/Multi-Level Storage – More Precise

    Why: Finer cell control may allow better gradation of resistance levels.

    Prediction: More stable analog states, improved signal-to-noise ratio for neuromorphic use.

    Use Case: Enhances suitability for in-memory computing or AI matrix operations.



    ---

    Summary Table

    Metric Expected Change @ 20nm Why

    Write Speed Faster (e.g. ~3.5 ns) Less capacitance, faster switching
    Power per Operation Lower Smaller area, lower voltage
    Endurance Same or better (~5–10B cycles) Less stress, if process control holds
    Retention Tunable but default may be lower Higher leakage, but voltage/time control compensates
    Analog Precision Higher Better control of smaller cell current/resistance states



    ---

    Bottom Line:

    Yes, if 4DS's process improvements from Platform Lot 5 are successful, Platform Lot 6 could show measurable performance gains across multiple metrics — especially speed, power, and analog precision. Retention might need adjustment, but their tunable architecture is well-suited to handle that.



 
Add to My Watchlist
What is My Watchlist?
A personalised tool to help users track selected stocks. Delivering real-time notifications on price updates, announcements, and performance stats on each to help make informed investment decisions.
(20min delay)
Last
2.6¢
Change
0.000(0.00%)
Mkt cap ! $51.52M
Open High Low Value Volume
2.6¢ 2.6¢ 2.5¢ $53.18K 2.053M

Buyers (Bids)

No. Vol. Price($)
32 6471699 2.5¢
 

Sellers (Offers)

Price($) Vol. No.
2.6¢ 406923 4
View Market Depth
Last trade - 11.54am 19/06/2025 (20 minute delay) ?
4DS (ASX) Chart
arrow-down-2 Created with Sketch. arrow-down-2 Created with Sketch.