Molecular electronics and single electron transistors. Molecular electronics: definition

Transcription

1 Molecular electronics and single electron transistors Single electron devices that function at room temperature require: Island sizes ~ 1 nm to have capacitances sufficiently small that k B T ~ 0.01 e 2 /C eq. Highly reproducible size and tunnel barriers. Natural thought: use single molecules as islands. History / motivation Nanotube SETs Recent single-molecule successes Molecular electronics: definition Molecules are used in bulk form in a number of prototype devices: nanotube field emission display Samsung. organic LED flexible display Universal Display Corp. Limit our discussion to situation where: Small number (< ~ 1000) of molecules per device Each molecule plays active role. 1

2 Origins of molecular electronics Aviram and Ratner (1974): Proposed using single molecule as a diode. electron-poor acceptor n-type analog O O CH 2 -CH 2 OCH 3 OCH 3 OCH 3 It seems to us reasonable to examine the potential use of molecules as components in electronic circuitry. electron-rich donor p-type analog Single molecule to replace macroscopic, active circuit element. The promise of molecular devices Chemical synthesis means producing ~10 23 components in parallel. Possibly low-cost. All molecules of type A should act exactly the same, with no statistical variations from molecule to molecule. Have designer toolkit over 100 years of synthetic chemistry background to draw upon, plus computational ability to predict energy levels and electronic structures. Molecules are intrinsically small, on a par with sizes of interest. 2

3 Big concepts Three key ideas that are fundamental departures from traditional electronics industry methodologies: Engineer single molecules to replace many components better than just miniaturization. single-molecule functionality self-assembly fault tolerance / adaptability Emulate biology use selfassembly and self-organization to avoid having to engineer assembly of components. Design architecture to handle large numbers of defective elements and adapt on-the-fly to component failures in-situ. Extremely small quantum dots Recall: Energy cost for changing occupation of quantum dots has two contributions: E c = Coulomb charging energy ~ e 2 /k B T classically E = Single-particle level spacing (ignoring interactions determined by solving Schroedinger equation with appropriate boundary conditions) In very small dots (L ~ wavelength of carriers), E can dominate. Result: spacing of Coulomb blockade peaks dominated by quantum confinement effects rather than classical Coulomb piece. 3

4 What are the tunneling barriers? Metal-molecule contacts are poorly understood! We know from our semiconductor work that interfacial charge transfer can drastically affect the character of the contact between dissimilar materials. Key ingredients: Energy level structures Relative Fermi levels What are the tunneling barriers? Resulting surface charge layer in metal can drastically alter work function (~ ev!). Transmission from metal to molecule depends critically on specific wavefunction overlap at individual contact atom. Best way for reproducibility probably chemical bonding (e.g. thiol group onto gold). Chemical bonding may not be good for establishing high conductance junctions, however. Example: STM on alkanethiols attached (covalently) to Au. Conduction is worse than vacuum (!). 4

5 Carbon nanotube SET Bezryadin et al., PRL 80, 4036 (1998). First attempts to study electronic properties of individual single-walled nanotubes resulted (unintentionally) in SET devices. Contacts were poor ( > 100 kω residue? Tube contamination? Poor coupling of tube modes to metal levels?) Bends broke tube into multiple segments. For long enough tubes, start out classical regime single-particle level spacing in tubes is small compared to charging energy. Carbon nanotube SET Recent development: Nanotube SET with very high charging energy (~ 45 mv!) by controlled deformation of tube. Buckle points lead to tunneling barriers. Short unbuckled region is island. Postma et al., Science 293, 76 (2001). 5

6 Other macromolecules Some recent work on making SETs using other macromolecules, particularly DNA. For example: DNA electronic conduction situation is a huge mess. Many contradictory results. Added issues of sample contamination from buffer solutions, carbonization under e-beam, irreversible damage, etc. Best assessment right now: DNA is a wide gap system even with no intentional disorder (poly-a/g rather than biological DNA), very unlikely to be a reasonable material by itself for SETs. Small organic molecules In last few years, first real development of techniques for making 3- terminal devices out of individual small molecules. First results (3 papers, + unpublished) have all shown Coulomb blockade features: contacts are sufficiently poor that there is not metallic conduction end-to-end in these systems. To do single-molecule three-terminal experiments, need controlled way of producing electrodes separated by nanoscale gaps (size of molecule). Main approach used: Electromigration from initially continuous metal strip. 6

7 How to make single-molecule devices Electromigration technique Park et al., APL 75, 301 (1999) current pulse Rely on chance that some fraction of devices will have a molecule positioned at the source-drain gap. Every device is different! 100 nm 30 nm Single-electron devices Electron-electron interactions can dominate transport properties. = single-particle level spacing = lowest energy of e-h excitation E c = Coulomb charging energy V S - + E E c Source Island Drain Source Drain V G - + 7

8 Single-electron devices For tunnel contacts and T <, E c, average charge island is well-defined. Near V D = 0, no current flow (to 1 st order) unless gate adjusts two charge states to be degenerate Coulomb blockade For typical small molecule,, E c ~ hundreds of mev! V S - + E E c Source Island Drain Source Drain V G - + Single-electron devices Data usually presented as differential conductance. Can overcome blockade with bias, too. V SD E I D n Source Drain V SD V G 8

9 Single-electron devices Data usually presented as differential conductance. Can overcome blockade with bias, too. V SD E I D Source Drain V SD n V G Single-electron devices Data usually presented as differential conductance. Gate shifts levels, discretely changes average island charge V SD E I D n Source Drain V SD V G 9

10 Single-electron devices Data usually presented as differential conductance. Gate shifts levels, discretely changes average island charge V SD E I D n n+1 Source Drain V SD V G Gate coupling concerns Electrostatics of this geometry favor poor gate couplings! Most of our devices were made on 200 nm thick SiO 2! metal (current source) AcS ~ 2 nm NO 2 NO 2 SAc metal (current drain) Naively, expect potential felt by molecule to be reduced from V G by a factor of ~ 200. dielectric gate electrode Best solution: use extremely thin gate oxide (e.g. native Al 2 O 3 ) Another subtle question remains. 10

11 How do you know what you have? Tunneling conductances depend exponentially on geometry. Statistical approach and systematic characterization are essential! Effects only present in samples with molecules? Charging energy sensible for a molecular-scale system? Charge states make sense in light of electrochemical data? Molecule-specific features? Example of a potential pitfall: metal blob from breaking process. V D [V] K Bare Metal V G [V] 20 IV curve zoology (C 60 devices) Total # of SMT electrode pairs examined = 724 1) No measurable currents. 30 % I D 2) Linear / weakly nonlin. 38 % 3) Strongly nonlin., no gate response. 19 % V D 4) Strongly nonlin., gateable. 13 % These are the ones we re interested in! 11

12 Vibrational effects I Another molecule-specific feature: Vibrational resonances E V SD hω Source Drain n n+1 V G 35 mev Vibrational effects I Another molecule-specific feature: Vibrational resonances E V SD Source Drain ev SD = hω n n+1 V G 35 mev 12

13 Vibrational effects I Another molecule-specific feature: Vibrational resonances E Source Drain Drain Voltage [V] mev Gate Voltage [V] C 60 device, 4.2 K C 60 SET First serious test: Try making 3-terminal molecular device using C60. Success rate ~ 10%. Park et al., Nature 407, 57 (2000). 13

14 C 60 SET Coulomb blockade diamond plot - changing charge on C60 by 1 unit. Coupling to gate varies from device to device. Signature of additional excitation with ~ 5 mev energy. C 60 SET Proposed explanation for that excitation: Vibrational mode of C60 bound to Au. 14

15 Higher order processes E In 2 nd order, can have virtual transitions. Virtual state violates energy conservation. V SD Source Drain n n+1 V G Higher order processes E In 2 nd order, can have virtual transitions. Virtual state violates energy conservation. V SD Source Drain n n+1 V G Nonzero conduction in blockade. Physicist: Elastic cotunneling Chemist: superexchange This is the off-resonant tunneling routinely seen in, e.g., alkyl SAMs. 15

16 Vibrational effects II Can also have virtual transitions that leave molecule excited. E V SD Source Drain n n+1 V G Vibrational effects II Can also have virtual transitions that leave molecule excited. E V SD Source Drain ev SD = hω n n+1 V G SET person: inelastic cotunneling Inelastic Electron Tunneling Spectroscopy! As shown, peak in 2 nd derivative. 16

17 Vibrational effects II Can also have virtual transitions that leave molecule excited. E Source Drain SET person: inelastic cotunneling Inelastic Electron Tunneling Spectroscopy! As shown, peak in 2 nd derivative. Higher order processes: spin E Virtual states allow antiferromagnetic exchange between conduction electrons in leads and molecule. V SD Source Drain n n+1 V G 17

18 Higher order processes: spin E Virtual states allow antiferromagnetic exchange between conduction electrons in leads and molecule. V SD Source Drain n n+1 V G Need unpaired spins on molecule. Result is formation of many-body Kondo state. At T = 0, unpaired spin is screened; resonant conduction maximized. Kondo regime Enhanced density of states at Fermi level of leads. Peak in conductance near zero bias V SD T K ~ ΓU exp( πε / Γ) deep in Kondo regime. Two ways of extracting T K : n n+1 0 ( SD 0, T) = s s ~ s T + 1/ TK G V G V G Zero bias resonance FWHM ~ 2k B T K /e for T << T K. Maximum G = 2e 2 /h for perfectly symmetric coupling to leads. 18

19 Kondo in SMTs: C 60 Bias [V] K 0 5E-8 1E-7 1.5E-7 2E-7 2.5E-7 3E Gate [V] Clear Kondo resonance, consistent with T K > 50 K. Kondo in SMTs: C K 0 Bias [Volts] E-6 5E-6 7.5E-6 1E E-5 1.5E Gate [Volts] Kondo resonance combined with ~ 35 mev parallel resonance. Inelastic tunneling process also enhanced by Kondo physics! 19

20 Biggest problems: Devices at this scale, made with not well-controlled methods, are dreadful to debug. Low yields of small gaps (10-15%) How can you tell what you have? No control of final molecular / metal configuration. Result: poor reproducibility. You can see just about anything once or twice. Room temperature diagnostics are tough: material stability, air exposure, inadequate diagnostic techniques. Best bet: use a designer molecule that allows you to clearly identify molecular conduction and distinguish it from other (unwanted) conduction mechanisms. Designer molecule example: Park et al., Nature 417, 722 (2002) Molecule is a metal complex between two linkers. Cobalt ion can exist in two charge states, one of which has even number of electrons (total spin 0), the other with an odd number of electrons. Plan: use magnetic field to figure out if they re really looking at the molecule. 20

21 Designer molecule example: Park et al., Nature 417, 722 (2002) Extra level appears because free spin on Co ion can be either aligned or antialigned with applied magnetic field. This level shifts linearly with applied field: Zeeman effect! Other evidence: Kondo effect. Kondo effect Park et al., Nature 417, 722 (2002) Presence of a localized spin that is free to flip can lead to a resonant enhancement of transmission. Result: A peak in conductance near zero bias. Peak grows logarithmically as temperature is lowered. Peak should Zeeman split in an applied magnetic field. 21

22 Designer molecule example: Liang et al., Nature 417, 725 (2002) New molecule includes two magnetic vanadium ions that can have either spin 0 or spin 1/2 depending on charge state. Plan: Total spin of molecule should vary as charge of molecule is varied. Designer molecule example: Kondo effect is again observed. This paper appeared backto-back in Nature with the previous paper we just discussed. The point of both: strong evidence that they are really looking at transport through a single molecule of interest. Liang et al., Nature 417, 725 (2002) 22

23 Kondo in transition metal complexes Can deliberately get Kondo physics by working with molecules that contain unpaired electrons! NCS 0.1 SCN Co2+ N N N N N N NCS VSD [V] Co Co SCN Ciszek et al., JACS, in press. I (µa) 50 0 Scan rate (V/s): Co +e => Co VG [V] 3+ Co -e => Co V vs. Ag AgNO3 Kondo in Co-complex 0.10 Nearly all successful SMTs made from this Co complex show Kondo peaks in one charge state. Can take data as a fn of T to watch peak height E VSD [V] Kondo temperatures depend on gate voltage in nontrivial way. 5K K K K E E E E E VG [V] 23

24 Robustness Devices can survive repeated thermal cycling to 300 K. V SD V G V G E-6 05E E-5 7.5E-6 1E-5 2.5E-5 1.5E E E-5 2E E-5 3E-5 2.5E-5 5E E-5 3E-5 4.5E-5 With better control of metal and surface chemistry, eventual technologies are not completely out of the question. Status so far Can make single-electron transistor devices with molecules as the active device region (island). Charging energies approaching / exceeding room temperature are possible. Manipulating single molecules and measuring conductance through them right now is extremely challenging, and has only been achieved definitively a very small number of times. Using the tools of chemistry, it is possible to construct molecules with properties that can serve as diagnostics, to confirm devices as single-molecule. 24

25 Prospects Good news: can get high charging energies. Bad news: does not really mitigate the other problems associated with single electron devices. In particular, low yields and poor reproducibility continue to hamper the work, even as a research tool. Good progress being made, however - five years ago we couldn t do any of this. Next time Molecular electronics beyond Coulomb blockade 25

26 Molecular electronics: beyond Coulomb blockade What kind of physics could lead to useful / interesting molecular devices, beyond the Coulomb blockade effects we ve been examining? Resonant tunneling diodes / transistors Switching / memory based on wavefunction engineering Switching / memory based on steric properties Progress in experiments A few basic techniques have been used to make most of the experimental progress in this area: Mechanical break junctions Nanopores Scanned probe microscopy Crossed wire structures We ll examine each of these in turn. 1

27 Mechanical break junctions Basic idea: Mechanical advantage: lateral motion from bending can be ~ 1000x smaller than vertical motion. Breaking can be done in UHV to avoid contamination. Bare metal tunneling current allows calibration of position. Images from van Wees group, Netherlands Zhou et al., APL 67, 1160 (1995). MBJ: Reed Start with unbroken wire, precoated with self-assembled monolayer (SAM) of interesting molecule. Break in UHV and allow molecules to rearrange. Bring junction back together for measurements, knowing what piezo voltage corresponds to the correct interelectrode spacing. Reed et al., Science 278, 252 (1997) 2

28 MBJ: Reed Reed et al., Science 278, 252 (1997) Results: Fairly reproducible IV curves that look like those at right. Differential conductance (blue trace) has features that can be identified with resonant tunneling through molecular levels. Theory must account for specific bonding of S to Au to get shape close. Theory still overestimates conduction by ~ 20x. di Ventra et al., PRL 84, 979 (2000) MBJ: Karlsruhe Reichert et al., PRL 88, (2002) Same basic plan, but this time do self-assembly on already-broken junction. Idea: compare symmetric molecule and asymmetric molecule cases, to see if the molecular structure is really what s determining conduction properties. 3

29 MBJ: Karlsruhe Reichert et al., PRL 88, (2002) Result: Asymmetric molecule always gives asymmetric IV traces. Symmetric molecule usually gives symmetric IV traces. MBJ: Karlsruhe Reichert et al., PRL 88, (2002) One other major point: Even with symmetric molecule, asymmetric curves can result. Moving electrodes while molecule is bonded either distorts bond angles at S-Au point, or rearranges last few Au atoms on surface. Confirms that molecular conduction characteristic depend critically on the atomic-scale details at the metal-molecule junctions. 4

30 MBJ: Crossed wires Kushmerick et al., PRL 89, (2002) Same point can be made in an even simpler crossed-wire geometry at room temperature. MBJ: Crossed wires and vibrations Kushmerick et al., Nano Lett. 4, 639 (2004) Junctions made this way are stable enough to allow measurements of d 2 I/dV 2. Inelastic electron tunneling features make sense. 5

31 MBJ: Gating Possible (though hard!) to make mechanical break junctions in gate-able configurations! Potentially very important technique. Champagne et al., cond-mat/ MBJ summary Highly productive research technique for examining single molecules. Shows that single molecule conduction is possible, though generally poor. Single molecule conduction depends crucially on atomic-scale details of bonding and metal surfaces. Gating is possible under certain circumstances. Endemic problems with technique: Mechanical stability essential. Very difficult to do temperature sweeps - everything moves due to differential contraction. This is a problem b/c standard way of deducing conduction mechanisms uses G(V,T). 6

32 Nanopores Images from Reed group, Yale Not a single molecule technique - more like ~ 1000 in parallel. Originally developed to study ~ nm diameter metal junctions. Nanopores Images from Reed group, Yale Requires self-assembly followed by evaporation of top electrode. Noone knows what interface looks like. Yield very very low (~ 1-2 %). 7

33 Nanopore devices: NDR Nanopore devices made with above molecule exhibit negative differential resistance, as shown. Mechanism not clear: temperature dependence is significant and steep. Proposed mechanism: detailed coupling of HOMO/LUMO to leads. Chen et al., Science 286, 1550 (1999) Donhauser et al., Science 293, 2303 (2001) Seminario et al., JACS 122, 3015 (2000) Nanopore devices: NDR With modification of molecule, can see NDR that persists up to room temperature. Other possible mechanism besides simple electronic structure: orientation of phenyl rings. 8

34 Nanopore devices: memory Reed et al., APL 78, 3735 (2001). Hysteretic IV curves. Initial ramp up: high conductance state. Ramp back down: low conductance state, until reset. Like NDR, effect is strongly temperature dependent. Requires very large electric fields to actuate. Nanopore devices: memory Can clearly use this kind of hysteretic IV curve as a memory! Again, potential mechanisms are either wavefunction oriented (shape of electron distribution in HOMO and LUMO) or steric (reorientation of side groups). 9

35 Nanopore devices: simple tunneling + vibrations Wang et al., PRB 68, (2003) Recent nanopore papers on very boring molecules show nice results. Conduction really is via tunneling. Inelastic features make sense. Wang et al., Nano Lett. 4, 643 (2004) Scanned probe microscopy: STM Another means of looking for interesting molecules and testing their (2-terminal) conductive properties is to use scanned probe microscopes. Allows highly controlled positioning of electrodes previously decorated with molecules. Can obtain I-V curves for single molecules. Can quickly examine many molecules Paul Weiss, PSU 10

36 Scanned probe microscopy: STM Downsides: Mechanical stability Difficult to do T-dep. measurements. Interpretation of contacts difficult - tunneling conductance is supposed to be proportional to product of local single-particle density of states of tip and surface. As result, tricky to deconvolve topography from electronic properties. One time it s not too hard: timevarying behavior Donhauser et al., Science 292, 2303 (2001). Scanned probe microscopy: AFM Conducting AFM is also a useful tool. Downside: contact area is typically significantly larger than single molecule. Upside: can vary contact force and see what happens. Surprisingly nice result at right: looks like as Au nanoparticle is pressed harder, more and more molecules are in contact in the circuit. Tricky to reconcile this with the data shown previously demonstrating that contact geometry is huge influence. Cui et al., Science 294, 571 (2001) 11

37 Nanopore and SPM summary Nanopores have produced some impressive results - NDR, memory w/ T-dependent measurements of ~ 1000 molecules in. Problem is, yields are terrible, and diagnostics on final molecule condition is essentially impossible. SPM techniques are much faster, and confirm some of the nanopore work (NDR, switching of conductance states). Unfortunately, SPM has its own set of experimental issues. Mechanisms behind these properties still argued. Crossed wire structures Collier et al., Science 285, 391 (1999) Collier et al., Science 289, 1172 (2000) On industrial side, HP and UCLA have been working on crossed wire structures. Suffer from same diagnostic concerns as nanopore devices, though much easier to fabricate (larger overlap). Molecules have hysteretic IV curves - may be used for nonvolatile memory. Limited to 2-terminal devices without gain, but that s not so crucial for nonvolatile memory schemes. 12

38 Crossed wire structures HP press release, 9/9/02 Combine nanoimprint lithography with this sort of device design. Result: 64-bit nonvolatile memory in ~(3 µm) 2, 10x higher density than commercially available NVRAM. Also built multiplexer for RAM readout. Predict possible marketready products w/in 5-10 years. Crossed wires: Big Problems! Problems have surfaced. Switching in a number of devices has nothing to do with molecules! People have suddenly gotten religion -- a nano researcher who shall remain nameless. Science, Oct. 24, 2003 Lau et al., Nano Lett. 4, 569 (2004) 13

39 Nanocell concept The researchers: Tour (Rice), Reed (Yale), Allara (Penn State), Mallouk (Penn State) Nanocell concept: Self-assemble large number of functionalized nanoparticles + molecules (e.g. NDR and memory molecules) across an array of leads. Training + field programming: apply voltage pulses; measure currents; train system to have desired functionality (!). Nanocell concept Every device is unique (analogous to detailed wiring of brain). Architecture that permits on-the-fly adaptation and training would be inherently robust + defect tolerant. A huge departure from standard electronics manufacturing practice! Molecular Electronics Corp. / Motorola paradigm: selfassembly + software architecture to allow training of circuits. input terminals output terminals 14

40 Summary and prospects Several research tools exist for trying to examine molecular conduction beyond Coulomb blockade. Basic qualitative features involving resonant tunneling through levels are understood. Details (contacts, quantitative results with predictive power) are still lacking. Basic physics responsible for certain interesting properties (NDR, switching) still under debate. We re a long way from practical molecular electronic devices, though applications in niche markets may be on the horizon. A fascinating field in which to be working! Organic electronics A thriving research (and commercial!) field already exists using molecular materials. Organic LEDs Organic FETs IBM 15

41 Organic electronics Key point: certain molecular materials act like semiconductors. Pentacene Ordered OSCs Small molecular crystals Polaron bands? conduction band valence band Disordered OSCs Glassy polymers Activated hopping? Regio-regular Poly (3-hexylthiophene) S S S S R R R R R = C 6H 13 localized states! Historical progress Mobilities have improved dramatically over the last 20 years. Some OLED devices are commercially available. However, many unanswered questions remain. IBM 16

42 Unanswered questions What are the basic physical parameters of carriers in organic semiconductors? (Effective masses, etc.) What is the physics of charge injection at OSC-metal interfaces? How does the conduction mechanism evolve from bands to hopping as disorder is increased? What are the fundamental limits of OSC performance? How important are electronic correlations in these materials? Next time. Basics of magnetism and magnetic materials. 17