robust resistive memory devices using solution ... · pdf filejani1,2, soumya sarkar1,2,...
Post on 25-Mar-2018
215 Views
Preview:
TRANSCRIPT
In the format provided by the authors and unedited.
1
Supplementary Materials
Robust Resistive Memory Devices Using Solution-
Processable Metal-Coordinated Azo-aromatics
Sreetosh Goswami1,2, Adam J. Matula3#, Santi P. Rath4#, Svante Hedström3,10#, Surajit Saha1,11,
Meenakshi Annamalai1, Debabrata Sengupta4, Abhijeet Patra1,2, Siddhartha Ghosh1, Hariom
Jani1,2, Soumya Sarkar1,2, Mallikarjuna Rao Motapothula1, Christian A. Nijhuis5,6, Jens Martin6,7,
Sreebrata Goswami*4, Victor S. Batista*3, T. Venkatesan*1,2,7,8,9
Affiliations:
1NUSNNI-NanoCore, National University of Singapore, Singapore 117576 2NUS Graduate School for Integrative Science and Engineering, National University of Singapore, Singapore 117456 3Department of Chemistry, Yale University, New Haven, CT 06520, U.S.A. 4Department of Inorganic Chemistry, Indian Association for the Cultivation of Science (IACS), Jadavpur, Kolkata-700032, India 5Department of Chemistry, National University of Singapore, Singapore 117543 6Centre for Advanced 2D Materials, National University of Singapore, Singapore 117546 7Department of Physics, National University of Singapore, Singapore 117542 8Department of Electrical and Computer Engineering, National University of Singapore, Singapore 117583 9Materials Science and Engineering Department, National University of Singapore, Singapore 117575 10Present address: Fysikum, Stockholm University, 10691 Stockholm, Sweden 11Present address: Department of Physics, Indian Institute of Science Education and Research (IISER), Bhopal, 462066, India
*Correspondence to: E-mail: venky@nus.edu.sg, victor.batista@yale.edu, icsg@iacs.res.in #These authors contributed equally to this work
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
SUPPLEMENTARY INFORMATIONDOI: 10.1038/NMAT5009
NATURE MATERIALS | www.nature.com/naturematerials 1
2
Section-S1: AFM (including c-AFM) measurement:
A JEOL JSPM 5200 AFM set up was used for all measurements. A CSC17/Ti-Pt tip (tip radius
of curvature < 30 nm, tip height: 20–25 µm, tip cone angle < 30°, resonant frequency ~12 kHz,
force constant ~0.15 N/m) was used for conducting-AFM characterization.
The c-AFM measurements were performed in contact mode in vacuum (at a pressure of 10−5
mbar). The cantilever probe was approached to the sample with a set point voltage, +2 V. J(V)
measurements were performed in contact mode and the resultant current was measured using a
pre-amplifier which can detect currents up to 5 µA with a 10 fA detection sensitivity.
Estimation of contact area between the tip and the sample:
The area of contact between an AFM tip and a sample can be estimated from several models
among which Johnson, Kendall and Roberts (JKR) and the Derjaguin-Muller-Toporov (DMT)
model can be taken as the respective upper and lower limits as indicated in the literature1, 2.
JKR Model:
According to JKR model, the relationship between the radius of contact (a) and the load force
(F) is given by the Hertz equation3,
𝑎 = #(%&'()#*+,' -)#*+,%&'(()#*+,),)/011
2 (1)
In order to calculate the radius of contact a, we need to estimate
1. The load force - Fl
2. Work of adhesion W12 between the tip and the surface
3. The effective elastic modulus between the probe tip and the sample - Eeff
4. Radius of the probe tip - R (<35nm from tip data sheet)
3
1. Load force - Fl:
We estimate the load force from the set point value (+2V) used during our measurement:
𝐹4 = 𝑘𝑆𝛿 (2)
where 𝑘 is the spring constant of the cantilever (= 0.18 N/m), S is the deflection sensitivity of the
cantilever and δ is the cantilever deflection (= 2V).
The deflection sensitivity (S = 62.5 nm/V) is extracted from the slope of the linear portion of the
attract part of the force curve obtained on a hard substrate viz. SrTiO3 (STO) as shown in Fig.
S9a.
2. Work of adhesion W12:
JKR model for pull-off force is expressed as3
𝐹89::;<== = 𝑐𝜋𝑅𝑊BC (3)
Where c is 1.5 for JKR model, R is the radius of the probe tip and W12 is the work of adhesion
Fpull-off is estimated using Eqn. 2 with same values of 𝑘 and S used to estimate the load force (F)
value. The δ = 1.5V is estimated from the F–d curve measured on our film surface (Fig. S9b).
By substituting the values of c, R and Fpull-off in Eqn. 3, W12 = 0.1023 N/m
3. The effective elastic modulus between the probe tip and the sample- Eeff
Eeff is calculated from the effective elastic modulus between two contacting bodies as shown in
Eqn. 4.
B/011
= (E(B;FGHI
,
/GHI+
B;FKLMIN0,
/KLMIN0) (4)
𝜗PQ8, 𝜗RST8:U, 𝐸PQ8 and 𝐸RST8:Uare the Poisson’s ratios and Young’s moduli of the probe tip and
sample, respectively. Substituting the values of Poisson’s ratio and Young’s modulus of the SiN
tip (𝐸PQ8 = 166 GPa, 𝜗PQ8 = 0.23) and sample (𝐸RST8:U = 25 GPa, 𝜗RST8:U = 0.4), we obtain, 𝐸U== =
4
34.039 GPa. We substituted these values in JKR model as shown in Eqn.1 to get a = 4.41 nm.
Therefore, the contact area 𝜋𝑎C = 60 nm2
DMT Model:
The expression for pull-off force as in Eqn. 3 is valid for the DMT model as well.4, 5 The DMT
model is an alternative model for contact mechanics and the contact profile is similar to the
Hertzian contact but it takes in to account additional attractive interactions outside the contact.
The area of contact by DMT model is given by,
𝑎 = #(%&'C)#*+,)/011
2 (5)
By substituting the values in Eqn. 5, we get a = 3.32 nm and contact area = 34.66 nm2
Although DMT predicts a lower value for the contact area, as a conservative choice we assume
an area of ~60 nm2.
We would like to highlight that as claimed in several reports6 the electrical area of contact is
estimated to be 10% of the physical contact. However, still to be on the conservative side, we
have used the area of 60 nm2 for the estimation of the current densities.
Precautions and control experiments:
Notably, as further precautions,
a. We measured the film topography before and after the I(V) measurements as presented in
Fig. S10a. We do not observe any blurring or artifacts due to tip damage.
b. We measured the I(V) several times on the same locations between measurements we
retracted and then approached the tip. Three such I(V) plots are presented in Fig. S10b
5
where after 1 we retracted and approached the tip again to measure 2, and using same
process after 2, we measured 3.
Energy Calculation in a c-AFM device: We calculated the switching energy using Es= IsVsts
where Is= on state current at switching voltage, Vs= switching voltage and ts= switching time (we
have taken 30ns which is the fastest we could demonstrate). For a 60nm2 device, Is= 5x10-7A,
Vs=0.1V, ts=30ns. Hence Es=(5x10-7 x 0.1 x 30x10-9) =1.5x10-15J= 1.5fJ.
Section-S2: Analysis of the in-situ Spectroscopy and Identification of Redox States of the
Film Molecules: The in situ spectroscopic techniques are used here to track the redox state of
the film molecules at individual values of applied bias voltage.
Interfacial Electron Injection: The material was isolated with all the ligands in the neutral
state, consistent with CV since only reduction peaks are obtained. However, once deposited on
ITO, the film shows a different spectral signature as shown in Fig. S11a,c. Interfacial charge
transfer from the electrode to the film is the cause of this change. The spectra we get on ITO
closely matches the 2-electron reduced UV-VIS spectra obtained in spectroelectrochemistry.
This electron transfer can be understood from the positioning of the energy levels of the film
molecules relative to the ITO electrode7 (see Fig. S11a). The film molecules have several low-
lying acceptor orbitals, including the LUMO and LUMO+1 which have lower or almost
equivalent energy values of energy (−4.42 eV and −4.15 eV) compared to the ITO work
function: F ~4.2 eV. Therefore, it is intuitive that the spin-casting of the molecules onto ITO
would result in a 2-electron reduction. This also explains the observations that such changes are
not observed when the film is deposited on Pt or Au bottom electrodes, which have larger F of
5.50 eV and 5.29 eV respectively.
6
However, it must be noted that the device works with Au, Pt bottom electrodes too. The
molecular acceptor energy levels are not initially accessible by metal Fermi-level. Hence unlike
ITO we do not start the cycle in the off-state. Instead, in the beginning we find the device in the
on-state. By application of a (negative) bias the electrostatic conditions between the metal and
the film allow occupation of higher redox states. Hence, in the first positive sweep, we stay in the
on-state but once we switch the device off, subsequently we still get similar J(V) like ITO bottom
electrode, only the voltage window of operation increases by approximately 10 %. The J(V)s
with Au and Pt bottom electrodes are shown in Fig. S12a,b.
Analysis of UV-Visible and Raman Spectral Trends: For the analysis of spectra it is
instructive to denote each of the redox states as listed below:
000: the state with all ligands neutral or unreduced
001: the state with one ligand singly reduced while the other two are neutral
011: the state with two ligands singly reduced while the remaining is neutral
111: the state with all ligands singly reduced
112: the state with two ligands singly reduced while one is doubly reduced.
We extract specific molecular redox states from Raman spectra based on the fact that for each
specific redox state there is a corresponding ratio of Raman integrated peak intensities (area
under the fitted Lorentzian). For example, in the (011) state 2/3 of all ligands are in the singly
reduced state, while 1/3 are neutral. This results into a ratio of 1:2 between Raman peaks at 1365
cm−1 (E1) and 1313 cm−1 (E2). In contrast, a 2:1 ratio between E1 and E2 peaks indicates the
molecular redox state (001). Notably in pure (000) and (111) phase, the area under the azo-
modes are quite similar which means that the oscillator strengths of these azo-modes in different
7
redox states are similar, as a result of which the sum of the three modes remain almost constant
(See Fig. 5b,d,f,h).
Since the Raman spectra are the integrated response from all film molecules, the assignment to
specific molecular redox-states is ambiguous. For example, a ratio of 2:1 between E1 and E2
peaks can be caused by all molecules in the (001)-state. Alternatively, 2/3 of the molecules may
be in the (000)-state and 1/3 of the molecules in the (001)-state. Most likely, in general, there is
some degree of mixture of different redox states.
Only in the on-state do we find very pure (000) and (111) redox states, supporting the claim that
high electrical conduction only occurs with all ligands of all molecules being in the same redox
state. Any mixture of ligand redox states within the molecule, or a mixture of molecules with
different redox-state within the film renders the device in an electrical off-state.
Note: We exclude the possibility of Ru2+/Ru3+ transition here for which we need to depopulate
HOMO (which is Ru-centred) of the complex. Both DFT and CV suggest a very low HOMO
energy of ~ −7.2 eV vs vacuum which can be attributed to pi-acidity of the azo ligands8, 9.
This argument is supported by our spectroscopic observations. It’s well known in literature and
suggested by DFT simulation that the Ru2+/Ru3+ transition would result in low energy absorption
peaks beyond 800 nm10, 11, 12, 13, 14 which we never observed in our voltage scan-range.
Here, we would also like to highlight that, one of the exclusivities of our molecular system is the
3-ligand centered low-lying molecular energy levels at −3.57, −4.15 and −4.42 eV. This is
clearly visible from the 3 well-resolved reversible redox peaks in a window of (0 to −1.1V w.r.t.
Ag/AgNO3). When compared to the other redox-active molecules used in memory devices, these
states are much deeper, as summarized in Table S4. These low-energy redox states are a
significant reason which renders stability, and robustness to our devices. In most other reports15,
8
16, 17, 18, in fact, the CV peaks appear to be irreversible which might be due to the lack of stability
of the redox states arising from their high-energy acceptor orbitals (see Table S4).
Charge Neutrality in the Device: Film molecules in different redox states essentially cause
charge build-up in the film which will be balanced by image charges on the electrode. Taking 3
extra electrons per molecule, the total charge density in the film is estimated to be on the order of
1020 cm−3 assuming a molecular density of 3.75×1019 cm−3 (obtained from Rutherford back-
scattering experiments). This value is neither prohibitive nor uncommon on the micro- or meso-
scale when it is compared to the charge densities reported on other devices such as organic field
effect transistors (OFET) and organic thermoelectric devices.19, 20, 21, 22 In fact, the depth of space
charge layer in Au and ITO to balance this charge build-up would be 0.0019 nm and 1.29 nm
respectively, which are perfectly reasonable values.
Reference: 1. ShiX,ZhaoY-P.Comparisonofvariousadhesioncontacttheoriesandtheinfluenceof
dimensionlessloadparameter.JournalofAdhesionScienceandTechnology2004,18(1):55-68.
2. HanJ,YeomJ,MensingG,JoeD,MaselRI,ShannonMA.Surfaceenergyapproachand
AFMverificationofthe(CF)ntreatedsurfaceeffectanditscorrelationwithadhesionreductioninmicrovalves.JournalofMicromechanicsandMicroengineering2009,19(8):085017.
3. JohnsonK,KendallK,RobertsA.Surfaceenergyandthecontactofelasticsolids.
ProceedingsoftheRoyalSocietyofLondonA:Mathematical,PhysicalandEngineeringSciences;1971:TheRoyalSociety;1971.p.301-313.
4. DerjaguinB,MullerV,ToporovYP.Effectofcontactdeformationsontheadhesionof
particles.ProgressinSurfaceScience1994,45(1-4):131-143.5. MullerV,DerjaguinB,ToporovYP.Ontwomethodsofcalculationoftheforceof
stickingofanelasticspheretoarigidplane.ColloidsandSurfaces1983,7(3):251-259.
9
6. CelanoU,HantschelT,GiammariaG,ChintalaRC,ConardT,BenderH,etal.Evaluation
oftheelectricalcontactareaincontact-modescanningprobemicroscopy.JournalofAppliedPhysics2015,117(21):214305.
7. LiY,ZhongH,LiR,ZhouY,YangC,LiY.High‐yieldfabricationandelectrochemical
characterizationoftetrapodalCdSe,CdTe,andCdSexTe1–xnanocrystals.AdvancedFunctionalMaterials2006,16(13):1705-1716.
8. ErnstSD,KaimW.Energyleveltailoringinruthenium(II)polyazinecomplexesbasedon
calculatedandexperimentalligandproperties.InorganicChemistry1989,28(8):1520-1528.
9. RillemaDP,AllenG,MeyerTJ,ConradD.Redoxpropertiesofruthenium(II)trischelate
complexescontainingtheligands2,2'-bipyrazine,2,2'-bipyridine,and2,2'-bipyrimidine.InorganicChemistry1983,22(11):1617-1622.
10. QiY,DesjardinsP,MengX,WangZ.Electrochromicrutheniumcomplexmaterialsfor
opticalattenuation.OpticalMaterials2003,21(1):255-263.11. KavanL,FrankO,KalbáčM,DunschL.SupramolecularAssemblyofSingle-Walled
CarbonNanotubeswithaRuthenium(II)−BipyridineComplex:AninSituRamanSpectroelectrochemicalStudy.TheJournalofPhysicalChemistryC2009,113(6):2611-2617.
12. WangS,LiX,XunS,WanX,WangZY.Near-infraredelectrochromicand
electroluminescentpolymerscontainingpendantrutheniumcomplexgroups.Macromolecules2006,39(22):7502-7507.
13. VickersSJ,WardMD.Facilepreparationofavisible-andnear-infrared-active
electrochromicfilmbydirectdepositionofarutheniumdioxolenecomplexonanITO/glasssurface.Electrochemistrycommunications2005,7(4):389-393.
14. LiX,NazeeruddinMK,ThelakkatM,BarnesPR,VilarR,DurrantJR.
SpectroelectrochemicalstudiesofholepercolationonfunctionalisednanocrystallineTiO2films:acomparisonoftwodifferentrutheniumcomplexes.PhysicalChemistryChemicalPhysics2011,13(4):1575-1584.
15. BandyopadhyayA,SahuS,HiguchiM.Tuningofnonvolatilebipolarmemristive
switchinginCo(III)polymerwithanextendedazoaromaticligand.JournaloftheAmericanChemicalSociety2011,133(5):1168-1171.
16. HuB,WangC,WangJ,GaoJ,WangK,WuJ,etal.Inorganic–organichybridpolymer
withmultipleredoxforhigh-densitydatastorage.ChemicalScience2014,5(9):3404.
10
17. SeoK,KonchenkoAV,LeeJ,BangGS,LeeH.Molecularconductanceswitch-onofsingle
rutheniumcomplexmolecules.JournaloftheAmericanChemicalSociety2008,130(8):2553-2559.
18. LeeJ,LeeE,KimS,BangGS,ShultzDA,SchmidtRD,etal.NitronylNitroxideRadicalsas
OrganicMemoryElementswithBothn‐andp‐TypeProperties.AngewandteChemieInternationalEdition2011,50(19):4414-4418.
19. SaekiA,SekiS.UnveilingChargeCarrierTransportinπ-ConjugatedMolecularWireon
Micro-andMacroscopicScales.ChemicalScienceofπ-ElectronSystems.Springer,2015,pp605-620.
20. XiaoY.Engineering,SynthesisandCharacterizationofNew-πConjugated(Macro)
molecularArchitecturesforOrganicOptoelectronics:applicationtowardambipolarmaterials.Paris6,2014.
21. KatzHE,PoehlerTO.InnovativeThermoelectricMaterials:Polymer,Nanostructureand
CompositeThermoelectrics.WorldScientific,2016.22. CampbellAJ,RawcliffeR,GuiteA,FariaJCD,MukherjeeA,McLachlanMA,etal.
Charge‐CarrierDensityIndependentMobilityinAmorphousFluorene‐TriarylamineCopolymers.AdvancedFunctionalMaterials2016,26(21):3720-3729.
23. KolthoffI.HandbookofAnalyticalChemistry.LouisMeites,Ed.McGraw-Hill,NewYork,
1963.Unpaged.Illus.$47.50.AmericanAssociationfortheAdvancementofScience;1963.
11
Figures: Table S1: Comparison with state of the art inorganic oxide memristors
Foot notes:
a Endurance of a device refers to how many times a device can be turned ON and OFF
b Stability refers to the temporal and temperature stability of a resistance state in a device, i.e. it
indicates the persistence of a written state.
c All resistances scaled to 100 nm2 size
12
Table S2: Comparison with state of the art (a) inorganic non-oxide and (b) organic memristors
13
Table S3: Device parameters: Device geometry and statistical distribution of electrical
characteristics are listed for all devices with planar and NP electrodes
14
Table S4: Comparison of molecular energy level with other redox active molecules used in resistive memory devices
a All potentials are referenced to Ag/AgNO3 (0.01M)23. b Pt used as Reference electrode.
15
Figure S1: Retention test and nonvolatility: After turning on Device-1 (device without NPs),
the voltage is withdrawn for an hour, after which the on-state is retained. This proves that the
device still retains its on state at voltages near 0 V, despite the existence of a non-conducting
plateau in this region.
Figure S2: Size dependent on/off ratio: The on/off ratio of devices of different top electrode
dimensions for (a) device-A (without NPs) and (b) device-B (with NPs). The error bars indicate
the standard deviations.
16
Figure S3: DFT: Raman stretching mode: (a) Correspondence between the experimentally
observed Raman modes to those obtained in DFT calculations (correspondence indicated via
arrows). (b) Formation of 2 isosbestic points by overlaying Raman spectra measured at 5
different voltage values (same as in Fig. 3e). (c) Schematic description of ligand redox-states.
17
a b
e
d c
f
-4V -1V
-4V -1V
-1V 1V
-1V 1V
1V 4V
1V 4V
g h
k
j i
l
4V 1V
4V 1V
1V -1V
1V -1V
-1V -4V
-1V -4V
Forward sweep: -4 è +4V
Reverse sweep: +4 è -4V
18
Figure S4: UV-vis in forward/reverse sweep: (a–f) UV-Vis and Raman spectra in forward
sweep corresponding to off-state (planar electrode). In the upper and lower panel, changes in
redox states can clearly be observed. (g–l) UV-Vis and Raman spectra in forward sweep
corresponding to the on-state. In contrast to the off-state, a clear charge transition can only be
identified in the central panel. Start: black, End: red.
19
Figure S5: Effect of counterion position on molecular orbitals: Isosurfaces (isovalue=0.035)
of HOMO, LUMO, LUMO+1, LUMO+2 of [Ru(L)3]2+ as structurally optimized in the presence
of one PF6− counterion in four different pockets. In all cases, the LUMO is largely localized on a
20
ligand far from the negatively charged PF6−, while the LUMO+1 and LUMO+2 are respectively
localized on the two remaining ligands.
Figure S6: Effect of counterion position on dipole moment: Dipole moment due to the spatial
separation between the molecule and the counterion. The dipole moment varies as a function of
counterion position.
21
Figure S7: Spectroscopic Characterizations: (a) 1H NMR spectrum of mer-[Ru(L)3](PF6)2 in
CD3CN solvent (*1,H2O;*2,solvent) (inset: aromatic proton resonances). (b,c) Segmented ESI-
MS spectra of mer-[Ru(L)3](PF6)2: (b) Masses observed from fragment C33H27N9Ru (z=2)
(i.e.[Ru(L)3]2+), (c) Masses observed from fragment C33H27N9F6PRu (z=1)(i.e.[Ru(L)3](PF6)+)
22
(blue lines show simulated mass spectra). (d) AFM image of spin-coated film, the r.m.s.
roughness is 1.03 nm. (e) Rutherford backscattering (RBS) spectrum of the [Ru(L)3](PF6)2 film.
Figure S8: Schematic of in-situ spectroscopy: Schematic presentation of in-situ (a) Raman and (b) UV-Vis spectroscopy.
23
Figure S9: Force–distance curve: An experimental force curve obtained on a (a) hard substrate
(STO) and (b) our film.
Figure S10: c-AFM measurement: (a) Film (in film/NP/ITO/YSZ structure, i.e. device-B)
topography before and after I(V) measurement in several points (indicated by white arrow). (b)
Three I(V)curves presents 3 different measurement at the same location; after measuring 1, we
retracted the tip and then approached to measure 2 and same process was once more repeated to
measure 3.
24
Figure S11: Charge transfer at electrodes: (a) The energy levels of the [Ru(L)3](PF6)2
molecule as obtained from cyclic voltammetry (CV) compared to various relevant known
electrode work functions. Notably LUMO and LUMO+1 are lower than or comparable to the
ITO work function, so these levels are likely to be populated with electrons from ITO, even
without applied bias. The HOMO energy is too low to contribute in electronic transport. (b) UV-
Vis-NIR spectrum of the film on ITO and in the isolated state of the compound. The isolated
state spectrum is retained on any insulating substrate. (c) Raman spectra of the film on ITO and
in the isolated state. (d) J(V) for different top electrodes. J(V)s are largely insensitive to work
function of the electrode materials excluding a Schottky-barrier.
25
Figure S12: J(V) for Au and Pt bottom electrode: J(V)s for (a) Au and (b) Pt bottom
electrodes where the first cycle is different from the subsequent cycles.
top related