![Page 1: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/1.jpg)
Magnetic tunnel junction beyond memory – from logic to neuromorphic computingWANJUN PARK
DEPT. OF ELECTRONIC ENGINEERING, HANYANG UNIVERSITY
![Page 2: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/2.jpg)
- 2 -
Magnetic Tunnel Junctions (MTJs)
• Structure
Free layer (FM 1)
Pinned layer (FM 2)
Tunneling barrier (Insulator: MgO)
Scalability Low energy High speed
High endurance CMOS compatibility
• Function Representation of non-volatile binary
state according to magnetization configuration
• Advantage
Bit line (BL)
Source line (SL)
Word line (WL)
AP or P state→ 0 or 1
1 Transistor + 1 MTJ
1GB STT-MRAM, Everspin (2016)
• High density memory
![Page 3: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/3.jpg)
- 3 -
Content
“ To find computing functions beyond memory from MTJ for integrated circuits”
MTJ Logic gates
Neuromorphic computing• Artificial MTJ neuron
• Artificial MTJ synapse
• Artificial Neurotransmission system
Construction of 2-input MTJ
![Page 4: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/4.jpg)
- 4 -
Basic configuration of MTJ
Methods for magnetization reversal of free layer
Anti-parallel state (RAP) Parallel state (RP)
FM 1
FM 2
FM 1
FM 2
• Single input configuration for switching to achieve the binary state
Switching variable 1: magnetic field switching (current induced Ampere field)
Switching variable 2: Spin-transfer torque (STT) switching (spin polarized current)
I
I
MTJ switching
![Page 5: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/5.jpg)
- 5 -
Two-input configuration of MTJ
• Motivation for multiple input extension
ISTT
IOF
RMTJ
Vin for ISTT
Hin by applying IAF
“ Increase of functional flexibility”- Reduction of switching stress by breakup of biases
- Increase of switching bias margin
• Physical variables for MTJ switching
- Ampere field-induce switching
- Spin-transfer torque (STT) switching
- Thermally assisted switching (TAS)
- Voltage-assisted switching (ME effect)
- Spin-orbit torque (SOT) switching
Multiple input is available for MTJ
Our choice: STT & Ampere field for two switching inputs Sharing integration methods developed for MRAM
X
IAF
Structure of 2-input MTJ
![Page 6: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/6.jpg)
- 6 -
Switching characteristics of 2-input MTJ
Size : (80 × 80) ~ (150 × 600) nm2
• MTJs for switching characteristics due to mixed inputs of STT & Ampere field
![Page 7: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/7.jpg)
- 7 -
Interpretation of 2-input switching
• Hs : Switching field• E0 : Energy barrier at zero magnetic field• aJ : Spin transfer torque• ac : Critical spin transfer torque • β = 2
• Case for switching to be P state
Eb+ = Eb0 - Eb
*
Required energy for STT switching
Energy due to Ampere field (Hext= Ha) :
Ha = 0 (at coercive center)Ha = 30 OeHa = -30 Oe
“Magnetic field assisted STT switching”
![Page 8: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/8.jpg)
- 8 -
NAND/NOR representation
Logic value Input 1 (V) Input 2 (H) MTJ OUT
1 VH HH RAP
0 VL HL RP
VL =0.2 V(<VS1), VH=0.3 V (VS1<VH<VS2)HL =-5 (Oe), HH= 5 (Oe)
VL =0.3 V (VS1<VL<VS2), VH=0.4 V (>VS2)HL =-5 (Oe), HH= 5 (Oe)
V H R
0 0 0
1 0 0
0 1 0
1 1 1
V H R
0 0 1
1 0 0
0 1 0
1 1 0
• Definition of binary states for each input • Assignment of input values to the STT input terminal
![Page 9: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/9.jpg)
- 9 -
All logical representation founded in MTJ
• 7 Boolean logic representations from possible 12 binary inputs of voltage biases for the STT switching
No.InitialState
Input set Input (V, H), Output (R) Logic function
VL VH HL HH (VL,HL) (VH,HL) (VL,HH) (VH,HH) RAP = 1, RP = 0 RAP = 0, RP = 1
1
RAP
0.2 0.3
–5 Oe 5 Oe
RAP RAP RAP RP NAND AND2 0.3 0.4 RAP RP RP RP NOR OR3 0.2 0.4 RAP RP RAP RP NOT V V4 0.2 0.2 RAP RAP RAP RAP TRUE FALSE5 0.3 0.3 RAP RAP RP RP NOT H H6 0.3 0.3 RP RP RP RP FALSE TRUE7
RP
–0.25 –0.35 RP RAP RP RP V NIMP H V IMP H8 –0.35 –0.45 RAP RAP RP RAP H IMP V H NIMP V9 –0.25 –0.45 RP RAP RP RAP V NIMP H V IMP H10 –0.25 –0.25 RP RP RP RP FALSE TRUE11 –0.35 –0.35 RAP RAP RP RP NOT H H12 –0.45 –0.45 RAP RAP RAP RAP TRUE FALSE
![Page 10: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/10.jpg)
- 10 -
MTJ Logic gate
• Logic gate for digital computing
Full schematic of a logic gate
- Cascading computing
![Page 11: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/11.jpg)
- 11 -
14 Boolean functions computed in MTJ logic gate
- XOR and XNOR are missing among full 16 Boolean logics
- Each function is confirmed by SPICE simulation modified with MTJ micro-model
![Page 12: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/12.jpg)
- 12 -
XOR/XNOR MTJ Logic gate
- XOR/XNOR gate could be completed by using “cascading computing”
• Conclusively, we have two types of MTJ logic gate which allow any digital computing
![Page 13: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/13.jpg)
- 13 -
Reconfigurable Logic
• Reconfigurability: further advantage of MTJ logic
< Truth table >
![Page 14: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/14.jpg)
- 14 -
Practical example for reconfigurable Logic
• Carry-out function with reconfigurable logic
< Truth table >
![Page 15: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/15.jpg)
- 15 -
Neurotransmission – spike signal carrying information
< Fast Adapting >
< Slowly Adapting>
Stimulus fortactile sense
SlowlyAdapting
FastAdapting
Pressure Frequency
![Page 16: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/16.jpg)
- 16 -
Neural coding
• Neural (Biological) coding
- Rate coding
Information spiking rate
- Time-to-first-spike, Phase Correlations, Spiking sequence, Synchrony etc.
Information Spiking pattern (Spike ordering in timing)
- Temporal coding
To find carrier for information according to “Strength and frequency “of input stimulus
![Page 17: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/17.jpg)
- 17 -
Telegraphic switching
Telegraphic switching by mixed effect of STT & Ampere field in 2-inpt MTJ
Total Energy (EM + ESTT) ~ ES
Toggling between AP and P stateStochastic characteristic Switching probability P(H, I) is
defined as the carried information
M. Pufall et al., Phys. Rev. B (2004)
0 10 20 30
10
20
30
40
R (kΩ
)
Time (s)
H = 80 Oe
−10 μA
−20 μA
10 μA
20 μA
I = -10 µA
![Page 18: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/18.jpg)
- 18 -
Rate coding
• Construction of neural coding : Rate coding
𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹(𝑩𝑩𝑩𝑩𝑩𝑩𝑩𝑩𝑩𝑩𝑩𝑩𝑩𝑩) = 𝒏𝒏𝒔𝒔𝒔𝒔(𝒏𝒏𝒏𝒏𝒏𝒏𝒏𝒏𝑹𝑹𝒏𝒏 𝑩𝑩𝒐𝒐 𝒔𝒔𝒔𝒔𝑩𝑩𝒔𝒔𝑹𝑹𝒔𝒔)𝑻𝑻 (𝑹𝑹𝑩𝑩𝒏𝒏𝑹𝑹𝒘𝒘𝑩𝑩𝒏𝒏𝒘𝒘𝑩𝑩𝒘𝒘
)exp(11),(
γβα +++=
IHIHPAP 𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝑴𝑴𝑻𝑻𝑴𝑴 = 𝐨𝐨𝐨𝐨
Information carrier : rateStimulus : H and I applied by independent inputs
)𝑹𝑹𝑨𝑨𝑨𝑨 (𝐨𝐨𝐨𝐨 𝑹𝑹𝑨𝑨)𝑻𝑻 (𝑹𝑹𝑩𝑩𝑹𝑹𝑹𝑹𝑩𝑩 𝒏𝒏𝑩𝑩𝑹𝑹
![Page 19: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/19.jpg)
- 19 -
Artificial neuron function
• MTJ-based neuron architecture representing the rate coding
![Page 20: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/20.jpg)
- 20 -
Neurotransmission
• Neuron system connected by “Synapse”
• Synapse• Neuron
# : 1014~15
103~4/neuron# : 1011~12
- Spike(=Action potential) generation(when the signal is above threshold)
- Neural coding Information is coded through spike train Rate coding
- Synaptic weight: plasticity for connection strength
- Weight modulation Potentiation / Depression Spike-timing-dependent plasticity (STDP)
(Spike-rate-dependent plasticity)
![Page 21: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/21.jpg)
- 21 -
Memristive character of MTJ
![Page 22: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/22.jpg)
- 22 -
MTJ-based artificial synapse
Δt > 0 Δt < 0
<Biological STDP>
R. Froemke et al., Nature (2002)
• Construction of input signal for “Spike-timing dependent plasticity”
![Page 23: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/23.jpg)
- 23 -
Artificial neurotransmission
• Neurotransmission system • Artificial neurotransmission system
• MTJs are commonly used for neural and synaptic functions
• Learning rule of “Spike-rate dependent plasticity” is possibly applied
![Page 24: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/24.jpg)
- 24 -
SRDP learning rule
![Page 25: Magnetic Tunnel Junction - CMU · Magnetic tunnel junction beyond memory – from logic to neuromorphic computing WANJUN PARK DEPT. OF ELECTRONIC ENGINEERING, HANYANGUNIVERSITY](https://reader036.vdocument.in/reader036/viewer/2022081613/5fb9300d300e33494d4eb0e8/html5/thumbnails/25.jpg)
- 25 -
Summary
MTJ was modified with two inputs for switching to achieve functional flexibility.
Then we found various computing functions for digital to neuromorphic computing.