1 / 24

STT-RAM Circuit Design

STT-RAM Circuit Design. MTJ Specs (Update), MTJ Sharing. I-STT MTJ Specs (Jianping). UPDATED SPECS R P ≈ 744 Ω TMR ≈ 136% AP→P: 630 μ A Max (breakdown current) 387 μ A for 3ns switching 330 μ A for 5ns switching P→AP: 1.5mA Max (breakdown current) Need more device measurements

azura
Download Presentation

STT-RAM Circuit Design

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. STT-RAM Circuit Design MTJ Specs (Update), MTJ Sharing

  2. I-STT MTJ Specs (Jianping) • UPDATED SPECS • RP ≈ 744Ω • TMR ≈ 136% • AP→P: • 630μA Max (breakdown current) • 387μA for 3ns switching • 330μA for 5ns switching • P→AP: • 1.5mA Max (breakdown current) • Need more device measurements • IWRITE(P→AP)/IWRITE(AP→P): 1.5-2 • 1ns read pulse (P→AP) with 1% chance of write: 220μA • AP→P might be better

  3. Maximum Write Currents (Thick Oxide) • Thin Oxide: 1.25nm • Max VDD = 1.0V • LMIN = 50nm • Medium Oxide: 2.2nm • Max VDD = 1.5V • LMIN = 100nm • Per μm width: IMAX,MEDIUM/IMAX,LVT = 93% • Thick Oxide: 5.2nm • Max VDD = 3.3V • LMIN = 230nm • Per μm width: IMAX,THICK/IMAX,LVT = 73%

  4. MTJ Sharing TMR Degradation (Reading)

  5. TMR Degradation SL BL<M> BL<2> BL<1> Parallel Resistance (R||) degrades TMR WL<2> WL<N> WL<1> Parasitic Parallel Resistance MTJ1,1 MTJ1,2 MTJ1,M MTJN,M MTJ2,1 MTJ2,2 MTJ2,M MTJN,2 MTJN,1

  6. Effective RP and RAP • Worst case TMR’: largest RP’ and smallest RAP’ • Largest RP’: • Smallest RAP’

  7. Effective TMR • Putting it all together: Example 1kbit Arrays: • TMR = 120%, M = 2, N = 16, 32-bit words: TMR’ = 4.8% • TMR = 120%, M = 2, N = 8, 64-bit words: TMR’ = 9.8% • TMR = 120%, M = 2, N = 4, 128-bit words: TMR’ = 20.7%

  8. Monte Carlo Simulations (M = 2, 3) • ERROR IN MATLAB CODE used to generate last week’s Monte Carlo plots → Not simulating the intended cases! • Fixed and reran simulations → results not so good • Ran multiple simulations with and w/o random variations in RP and TMR (based on worst case from Jianping) • Extracted worst case TMR’ and TMR’ for 10% read error • For 128-bit words, with bit read error = 10.0%: • # error correcting bits = 36 (1/5 word) • Probability of a word error: 1 in 6,788 reads • # error correcting bits = 32 (1/4 word) • Probability of a word error: 1 in 3.59x106 reads • # error correcting bits = 43 (1/3 word) • Probability of a word error: 1 in 13.2x1012 reads

  9. TMR’ vs. N for 1T-2MTJ (M = 2)

  10. Monte Carlo Simulations: M = 2, N = 4 • TMR = 120% • RP = 500Ω • 25k Simulations • TMR’ • Worst Case = 20.7% • ~10% Read Error = 30.0% • TMR = 120%, 3σ = ±12% • RP = 500Ω, 3σ = ±50Ω • 25k Simulations • TMR’ • Worst Case = 3.7% • ~10% Read Error = 25.4%

  11. Monte Carlo Simulations: M = 2, N = 8 • TMR = 120% • RP = 500Ω • 25k Simulations • TMR’ • Worst Case = 9.8% • ~10% Read Error = 14.6% • TMR = 120%, 3σ = ±12% • RP = 500Ω, 3σ = ±50Ω • 25k Simulations • TMR’ • Worst Case = -4.1% • ~10% Read Error = 12.4%

  12. Monte Carlo Simulations: M = 2, N = 16 • TMR = 120% • RP = 500Ω • 25k Simulations • TMR’ • Worst Case = 5.1% • ~10% Read Error = 8.5% • TMR = 120%, 3σ = ±12% • RP = 500Ω, 3σ = ±50Ω • 25k Simulations • TMR’ • Worst Case = -10.3% • ~10% Read Error = 4.7%

  13. TMR’ vs. N for 1T-3MTJ (M = 3)

  14. Monte Carlo Simulations: M = 3, N = 4 • TMR = 120% • RP = 500Ω • 25k Simulations • TMR’ • Worst Case = 0.0% • ~10% Read Error = 15.6% • TMR = 120%, 3σ = ±12% • RP = 500Ω, 3σ = ±50Ω • 25k Simulations • TMR’ • Worst Case = -10.1% • ~10% Read Error = 15.9%

  15. Monte Carlo Simulations: M = 3, N = 8 • TMR = 120% • RP = 500Ω • 25k Simulations • TMR’ • Worst Case = -11.2% • ~10% Read Error = -0.6% • TMR = 120%, 3σ = ±12% • RP = 500Ω, 3σ = ±50Ω • 25k Simulations • TMR’ • Worst Case = -16.7% • ~10% Read Error = -1.0%

  16. Monte Carlo Simulations: M = 3, N = 16 • TMR = 120% • RP = 500Ω • 25k Simulations • TMR’ • Worst Case = -15.9% • ~10% Read Error = -11.2% • TMR = 120%, 3σ = ±12% • RP = 500Ω, 3σ = ±50Ω • 25k Simulations • TMR’ • Worst Case = -23.4% • ~10% Read Error = -11.4%

  17. MTJ Sharing Device IREAD/IWRITE Requirements (Writing)

  18. Defining IREAD,MAX & IWRITE,MIN • IREAD,MAX: The maximum read current such that the probability of flipping the MTJ is less than some ε(i.e. ε = 0.1% → IREAD,MAX = 200μA) • IWRITE,MIN: The minimum write current such that the probability of failing to flip the MTJ is less than some ξ(i.e. ξ = 0.1% → IWRITE,MIN = 600μA)

  19. IREAD/IWRITE for 1T-2MTJ & 1T-3MTJ BL<1> • Example: 1T-2MTJ architecture WL<2> WL<1> BL<2> IWRITE,MIN α∙IREAD,MAX

  20. IREAD/IWRITE for 1T-2MTJ RP Case 1: RAP Case 1: RP Case 2: RAP Case 2: “1” “1” “1” “1” “0” “0” “0” “0” RAP RP RAP RP RP RP RP RAP RP RAP RP RAP RP RP RAP RP

  21. IREAD/IWRITE vs. TMR for 1T-2MTJ

  22. IREAD/IWRITE for 1T-3MTJ RP Case 1: RAP Case 1: RP Case 2: RAP Case 1: RP RP “1” “0” RAP RP RAP RP RP “1” “0” RP RP RP RAP RAP RP RAP “1” “1” “0” “0” RP RP RAP RAP RP RP

  23. IREAD/IWRITE vs. TMR for 1T-3MTJ

  24. SUMMARY • TMR Degradation (READING) • M = 2: • Not as good as previously thought • Read circuit need to work for 25-30% TMR • 10-12% for more wordlines • M = 3: not really possible (negative TMR) • IREAD/IWRITE (WRITING) • For TMR = 120%, χ = 1.5-2: • M = 2: IREAD/IWRITE > 0.36-0.43 • M = 3: IREAD/IWRITE > 0.42-0.49

More Related