ANALYSIS AND MODELING OF 1/f NOISE IN IGZO TFTS Gerard Uriarte, Wondwosen E. Muhea, Benjamin Iñiguez Dep. of Electronic Engineering, University Rovira i Virgili, Tarragona (Spain) Thomas Gneiting AdMOS GmbH Frickenhausen (Germany)
Outline Introduction about the origin of Flicker noise Results of 1/f noise analysis and modeling in IGZO TFTs Bias and channel length dependence Noise mechanism Conclusions 2
OBJETIVE Determine the cause and origin of the noise in IGZO TFTs Determine the behavior of noise with V G L 3
1/f NOISE Study of the scalability of noise Study of the origin of 1/f noise Dependence on : VG Channel length 2 theories Mobility fluctuations, Δµ Carrier number fluctuation, ΔN 4
Origin of the 1/f noise Carrier number fluctuation, ΔN McWhorter Model, 1957 Charge trapping and release Generation near insulating interface Causes 1/f noise IGZO TFT, source www.semind.com 5
Origin of 1/f noise Carrier number fluctuation, ΔN S ID = k /f µ/ C ox L 2 I D V DS / (V GS V th ) (1) Dimitradis et al proposed, for poly-si TFTs: (2) Therefore : (3) 6
Origin of 1/f noise Mobility fluctuations, Δµ Hooge model, 1969 Fluctuations of carriers mobility in the conduction channel. 7
Origin of 1/f noise Mobility fluctuations, Δµ Hooge proposed (4) (5) N el number of carriers: (6) Therefore: (7) 8
Devices Analized IGZO TFT TNO, HOLST CENTER, The Netherlands A total of five IGZO TFT were analyzed, all of them with a 100µm channel width and 15, 20, 30 and 50µm channel length, for a better scalability comparison. All the devices have a bottom gate, an insulation thickness of 200nm and an ESL thickness of 100nm, with 5µm design rules. 9
Origin of 1/f noise Model by Ghibaudo 1991, Δµ-ΔN Combination of both theories Proposed: For a pure ΔN model, the term with α s is absent As a result, if S ID (f)/ I D 2 and [gm/ I D ] 2 as a function of I D are parallel, it is because the fluctuation of the number of carriers. 10
Origin of 1/f noise Summary: Mc Whorter Model Δµ Hooge Model ΔN Ghibaudo Model Δµ-ΔN Slopes m = -2 Proportionality f 1 m = -1 m = -1 Scalability L SID VG SID 11
Results, Bias dependence IGZO TFTs follow, 1/ f γ, with γ = 1. 10 7 10 8 10 9 10 10 10 11 10 12 10 13 10 14 10 15 10 16 1 10 100 10 3 10 4 Hz Normalized noise drain current spectral density ( S ID / I D 2 ) for all targeted devices, measured at the 4, 12 and 20V gate voltage ( V GS Vth) in the linear region of device operation ( V DS = 1.0 V). W and L in µm. 12
Results Bias dependence Dependence on VG. TFT IGZO 10 7 10 8 10 9 10 10 10 11 10 12 10 13 1 10 100 Vg S ID (f)/ I D 2 versus V GS (symbols), their linear approximations (solid lines) and extracted slope values (inset) at 12 Hz for different (V GS V th ) values, with V DS = 1 V. W and L in µm. 13
Results, Bias dependence W L ( V V ) V ( 1+ λ V ) GS T DS DS I DS = Ci µ FET 1 m m W V DS 1+ R Ci µ FET ( VGS VT ) 1 + L S ( VGS VT ) α µ µ γ FET = o ( V ) γ GS VT V AA 14
Results Mobility is a power law of ( V GS V th ) in IGZO TFTs, as well as I DS at low V DS Extraction of threshold voltaje in IGZO TFTs H( V GS )= 0 V GS I D (x)dx / I DS ( V GS ) = 1/2+γ ( V GS V th ) 6 H Linear regression H(Vg) 4 2 0 0 5 10 15 20 Vg 15
Results, Channel length dependence 10 8 Dependence on L. TFT IGZO S ID / I D 2 versus gate length at different (V GS V th ), with V DS = 1 V, and f =12 Hz, for IGZO TFTs in 15-µm, 20-µm, 30-µm, and 50-µm gatelength devices. 10 9 V GS = 4V 10 10 V GS = 12V 10 11 V GS = 20V 10 12 10 100 L, um Agrees with theory, Vg L noise Noise produced in channel region. 16
10 7 10 8 10 9 10 10 RESULTS, Origin of noise S ID (f)/ I D 2 and [gm/ I D ] 2 are parallel as a function of I D. Device, 15L 100 10 1 0.1 10 7 10 8 10 9 10 10 Device, 20L 100 10 1 0.1 10 11 10 12 10 13 10 14 0.01 10 3 10 4 1 10 5 10 7 10 6 10 5 10 4 10 3 Id, A 10 11 10 12 10 13 10 14 0.01 10 3 10 4 1 10 5 10 7 10 6 10 5 10 4 10 3 Id, A This means that the main origin of noise is the fluctuation of the number of carriers 17
RESULTS, Origin of Noise Log log plot of the normalized noise drain current spectral density ( S ID (f)/ I D 2 ) versus gate voltage overdrive ( V GS V th ) 10 5 10 6 10 7 slope-1 10 8 10 9 10 10 10 11 10 12 10 13 slope-2 10 14 0.1 1 10 100 Vgs-Vth, V Slope closer to -2. This means that the main origin of noise is the fluctuation of the number of carriers 18
CONCLUSIONS Channel length and bias dependences: L increases, S ID (f)/ I D 2 decreases VG increases, S ID (f)/ I D 2 decreases ~ (1/ f γ ), γ = 1 1/f Noise: TFT IGZO, Fluctuation of the number of carriers, N. Log ( S ID (f)/ I D 2 )-log( V GS V th ) Slope -2. Caused by the mechanism of trapping and reléase at the interface of the insulating layer. 19
8th SINANO Modeling Summer School Tarragona (Spain), September 25-28 2018 23 lectures targeting different techniques of device modeling, characterization and application to circuits. Among the technologies targeted: SOI CMOS, Multi-Gate MOSFETs, GaN HEMTs, Organic TFTs, IGZO TFTs, Organic Photovoltaics, Quantum Computing devices, Synaptic devices, Reduced registration fee until September 12. Cheap fee especially for students. fundacio.urv.cat/congresos/sinano-modelling-summer-school Looking forward to seeing you in Tarragona! 20
DOMINO RISE PROJECT Goal: compact models for Amorphous Oxide Semiconductor and Organic TFTs integrated in design tools 4 universities, 2 research centers (foundries), 3 EDA companies Coordinator: URV Dec 1 2014-Nov 30 2018 21
THANKS FOR YOUR ATTENTION??? 22