PCM memories: an overview

Transcription

1 PCM memories: an overview di e Politecnico di Milano & IUNET Nov. 26, 2008 Outline PCM concept and basic operation Modeling Switching Set/Reset programming Cell structures and scaling perspectives Reliability issues Retention, Drift Conclusions Page 1

2 Phase Change Memory 1970 Die: 122 mil X 131 mil Capacity: 256 bits Reset: <200 ma,< 25V, 5 ms Set: 5 ma, ~ 25V, 10 ms Read: 2.5 ma, < 5V Nonvolatile and Reprogrammable, the Read-Mostly Memory is Here, R. G. Neale, D. L. Nelson, and Gordon E. Moore, Electronics (Sept. 1970) p Chalcogenide Alloys IVA VA VIA VIIA 1970 Xerography C N O F Si P S Cl Ge As Se Br Sn Sb Te I Pb Bi Po At 1990 DVD-RW CD-RW 2000 Memories Chalcogens OUM (Ovonic Universal Memory) PCM (Phase Change Memory) PRAM (Phase-change RAM) Page 2

3 Storage Element Top electrode Active region Resistor Crystalline GST (Ge 2 Sb 2 Te 5 ) Bottom electrode Sensing Mechanism Set 1 I Crystal Low resistance V Reset 0 Amorphous High resistance Page 3

4 SET to RESET Current Set Time Reset RESET to SET Current Reset I x Time Set Page 4

5 IV curve 0.75 Current [ma] Crystal Amorphous V th Voltage [V] Key effect for low voltage operation Threshold field (V th /d) The lower the threshold field the better Programming Curve Resisstance [Ω] Crystal Amorphous Set Programming Current [ma] Reset Page 5

6 Resistance loss 1) 1) 2) 2) 3) 3) Temperature dependence years Crystallization Time [s] C /k B T [ev -1 ] Page 6

7 Material selection rules Memory window Reset Power Retention resistivity of amorphous phase resistivity of crystalline phase melting temperature crystallization temperature activation energy high (0.3 MOhm.μm) high (350 Ohm.μm) low (621 o C) high (155 o C) high ( eV) Bias voltage threshold field low (30-40 V/μm) Material options Sb 2 Te 3 -GeTe GeTe-InTe higher amorphous stability, better retention higher maximum operating temperatures slower O/N doped GST higher resistivity lower current Doped SbTe, GeSb fast growth materials Page 7

8 Summary Beyond the capacitive concept Fast write (set/reset) and read Medium/low voltage write Supply (threshold field) Power consumption (melting temperature) Retention (cristallization temperature, activation energy) Window (resistivity change) Outline PCM concept and basic operation Modeling Switching Set/Reset programming Cell structures and scaling perspectives Reliability issues Retention, Drift Conclusions Page 8

9 PCM Physics and Modeling I-V curves of both phases Cell programming Threshold switching Self-heating melting/quenching a-x transition Electro thermal & phase change model Set to reset transition Current I R Time Page 9

10 Electro-thermal thermal simulation I=400 μa GST 50 nm SiO2 Heater SiO2 I R Electro-thermal thermal simulation I=500 μa GST 50 nm SiO2 Heater SiO2 I R Page 10

11 Electro-thermal thermal simulation I=600 μa GST 2 GST SiO Heater SiO2 50 nm I R Electro-thermal thermal simulation I=800 μa GST GST SiO Heater SiO nm I R Page 11

12 Electro-thermal thermal simulation 2 GST Amorphous GST SiO Heater SiO2 50 nm I R Programming curve I R Time Resistance [Ω] lance μ-trench R set Current [μa] Page 12

13 Material parameters ρ=17mωcm κ= 1W/Km ρ=50ωcm κ= 0.3W/Km + Wiedemann-Franz + Thompson ρ=25mωcm κ= 12W/Km κ= 0.7W/Km Reset to set transition Current I x Time Page 13

14 Reset to set transition Current I x Time Reset to set transition Current I x Time conduction starts in narrow filaments crystallization takes place in the same regions Page 14

15 Reset to set transition Current I x Time Reset to set transition Current I x Time Page 15

16 Reset to set transition Current I x Time Reset to set transition Current I x Time Page 16

17 Reset to set transition Current I x Time Set state: set-time time dependence 10 6 Resistance [Ω] ns 200 ns 1 μs Programming current [μa] Page 17

18 Coupling with crystallization Electro-thermal simulation Δt T i Nucleation & Growth Transient loop Updated phase distribution Nucleation and Growth implemented through a Montecarlo method Crystallization Thermal enhancement of atomic mobility drop of free energy gain per unit volume Nucleation - growth Probability density [ns -1 ] 10 0 Growth Nucleation x Temperature [ C] Page 18

19 Summary Numerical tools have been developed to support the design and optimization of PCM technology Carrier transport and heat flow are self-consistently coupled in the frame of semiconductor device simulator Material parameters have been tailored to quantitatively account for experimental results These tools are essential to design further scaled cell architectures Outline PCM concept and basic operation Modeling Switching Set/Reset programming Cell structures and scaling perspectives Reliability issues Retention, Drift Conclusions Page 19

20 PCM and Joule heating Endurance Cross talk Power dissipation / current consumption 1,8V Reset current 500μA/cell Write cycle time 1μs 5V 20 cells in parallel 1Mcycles/sec 16Mb/s = 2MB/s Current reduction by scaling Current [ma] Crystal Amorphous Crystallization Melting Electronic switching 680 C C Voltage [V] Temperature 1 k ΔT M = P d. R TH 1/k k P d = R. I 2 1/k Page 20

21 PCM scaling strategies Parameters isotropic anisotropic GST/Heater contact area A cell 1/k 2 1/k 2 Layer thickness Electrical/Thermal Resistances Power dissipation Current Voltage Current density R P cell I V cell J 1/k k 1/k 1/k 1 k 1 k 2 1/k 2 1/k R set x =const Current reduction by scaling by material engineering (active/heater) by cell architecture Page 21

22 Lance / Ring (A) (B) Hori et al. VLSI 2005 Ahn et al. VLSI Pillar / Pore (A) (B) Page 22

23 μtrench Metal GST Heater F. Pellizzer et al, VLSI 06 Contact area: heater thickness x sublitho Technology benchmarking F [nm] STM-Intel VLSI Samsung IEDM Hynix J. Sem. Sci. And Tech. 8(2), 128, Hitachi IEDM06 ISSCC IBM- Qimonda- Macronix VLSI IBM- Qimonda- Macronix VLSI Cell type μtrench/ lance, bipolar, Ring, GST N-doped, diode, 5,8F 2 - Lance, GST -Ta 2 O 5 layer, MOSFET Pillar, GST N-doped MOSFET Pore, GST N-doped MOSFET Ireset [ma] 0.4/ Array size [# bit] 128M 512M 512M 4M Mini array 256k Set time [ns] (MLC) 80 Endurance [#] > Page 23

24 From MOSFET s to diodes Samsung, VLSI 04 Samsung, IEDM Summary Among the various candidates for future NVMs, PCM technology has shown the most convincing prospects for commercial products High density arrays as well as embedded solutions have been successfully demonstrated Short-term: valuable solution for embedded applications, code storage, high performance memory systems Long-term: solution for data storage applications Page 24

25 Outline PCM concept and basic operation Modeling Switching Set/Reset programming Cell structures and scaling perspectives Reliability issues Retention, Drift Conclusions Data-loss statistics Repetitive measurements on the same cell show statistical spread data loss is erratic Page 25

26 Weibull distribution of t fail Failure time shows a Weibull distribution Consistent with failure due to percolation mechanism T impact on statistics Statistical spread increases for increasing T: why? Page 26

27 Extraction of average grain size Cumulative distrib. % T=210 o C 190 o C 180 o C 1 r C =2.2nm 3.3nm 5.6nm vg t r C = rn Retention lifetime [s] X Data-retention projection T fail from 105 C (typical) to 102 C (worst over 1Gbit) Page 27

28 Retention at the nanoscale T x [ o C] Si nanodot [1] GeSb nanodot [2] 200 GeSbTe nanowire [3] D [nm] Both cristallization time and activation energy decrease at the nanoscale Impact of surface N/G Tailoring material composition [1] Hirasawa et al., APL 06, [2] Raoux et al., EPCOS 07, [3] Lee et al., Nat. Nanotech., Reliability issues Reliability issue Crystallization Structural relaxation Impact on Cell Resistance decrease Resistance increase Cycling endurance Program disturb Read disturb Stuck set/reset Resistance decrease/increase Switching and resistance decrease Page 28

29 Resistance drift 10 7 ν r =0.12 Resistance [Ω] ν r =0.1 ν r =0.05 ν r = Time [s] SR physical characterization 1. Calorimetry exothermic reaction in a-si, a-ge and chalcogenide glasses (S. Roorda et al., PRB 1991) 2. Increase of viscosity with time (J. A. Mullin, PhD thesis, 2000) 3. Photoconductivity in a-se indicate a power-law decrease of trap density (K. Koughia et al., JAP 2005) Page 29

30 Poole-Frenkel conduction Δφ Less defects Lower current Higher resistance Current [A] Initial After bake (120C, 1D) Calculated Voltage [V] Ielmini et al., IEDM Kinetic model for SR SR is due to atomic rearrangements in the disordered structure Energy E A τ = τ 0 e E A k T B kinetic reproduced by metastable defect relaxation with broad distribution of activation energies Material engineering Reaction coordinate Programming/reading schemes to minimize the effect Page 30

31 Summary Perspectives of PCM at the nanoscale calls for optimization/engineering of material properties Drift and defects dynamics, intrinsically linked to the amorphous phase has to faced and dominated to make reliable MLC. Great fun in the next future! Acknowledgments STMicroelectronics, Intel, Numonyx EU (CAMELS) D. Ielmini, A. Pirovano, A. Redaelli, D. Mantegazza, U. Russo and many other students of the Nano Lab Politecnico di Milano Page 31