Electron and Spin Transport in Graphene and Metallic Channels with Photosystem I Monolayers

(1)

Electron and Spin Transport in Graphene and Metallic Channels with Photosystem I Monolayers

Master research - Applied Physics July 25, 2016

T B

2041065

Group: Physics of Quantum Devices

Daily supervisor: X. Y

Supervisor: Prof. C.H. W

Second examiner: Dr. L.J.A. K

(2)

Abstract

In this research monolayers of photosystem I (PSI) and their charge transfer characteris cs are inves gated. We have self-assembled them onto chemical vapor deposited graphene sheets via pep de linkers, which should also orient the PSI to have its dipole downwards. Tapping mode atomic force microscopy (AFM) as well as conduc ve probe AFM are employed to confirm the coverage and orienta on of the PSI complexes. These monolayers are also integrated in a backgated graphene field-effect transistor in order to study the electrical proper es of graphene before and a er self-assembly. Illumi- na on experiments have been performed to observe a varia on in doping level of the graphene induced by a changing dipole in PSI. In addi on, a paper introducing spin selec vity in charge transfer by photosystem I is analyzed regarding reproducibility. Calcu- la ons based on a resistor model are performed. These imply that the measured signal requires larger currents than known charge transfer rates for PSI predict. Finally, we introduce a conceptual device based on a nonlocal spin valve geometry to measure the signal from spin selec vity in PSI.

(3)

Chapter 1 Introduc on

To date, photosynthesis remains a very intriguing energy harves ng mechanism, as nature has developed its own biological solar cell technology. It is commonly known that this process occurs in the green parts of plants. Specifically, it happens within the chloroplasts residing in the plant cells, but also within cyanobacteria. These organelles contain so-called thylakoids (fig. 1.1a). It is on this thylakoid’s membrane where the most interes ng charge transport for photosynthesis occurs (fig. 1.1b). The incident light is harvested by charge transfer in two complex protein systems:

Photosystem I (PSI) and Photosystem II (PSII). Their number relates to the order of discovery [1].

The research presented in this report aims exclusively at studying PSI from cyanobacteria. The internal quantum eﬃciency within this system is nearly 100% [2], meaning that almost all of the photons absorbed in PSI result in charge transfer. However, the overall eﬃciency of the complete photosynthesis chain lays somewhere around 1% [3].

(a) (b)

Figure 1.1: (a): Schema c of the chloroplast structure. Within the stroma liquid there is a granum structure, made up from stacked thylakoids. Photosynthesis occurs in the thylakoid lumen. Figure from [4]. (b): Chain of photosynthesis processes occurring within the thylakoid membrane. PSII and PSI are the consecu ve light harves ng systems, driving charge transfer through the membrane. Ul mately, this leads to the forma on of ATP and NADPH energy carriers.

Figure from [5].

Therefore, implemen ng this photosystem in new devices has a racted a lot of a en on. For example, it has been used as ac ve layer in diﬀerent types biohybrid photovoltaic cells [6, 7]. In addi on, charge transfer dynamics in PSI are being studied to gain more understanding of such

(6)

complex organic systems. It is known that PSI has an internal dipole of 1000 D¹[8], which points upwards in ﬁgure 1.2a. The electron transfer path upon photo-excita on is depicted in ﬁgure 1.2b. Incident light excites the P700 reac on center and an excited electron then travels towards the iron-sulfur clusters (Fx, Faand Fb) via chlorophyll a (A0) and a phylloquinone molecule (A1) [9].

From here it will par cipate in the rest of the photosynthe c chain.

(a) (b)

Figure 1.2: (a): Detailed structure of a PSI monomer from plant cells, slight diﬀerences exist for bacterial PSI. The P700 reac on center and iron-sulfur clusters, (Fx, Faand Fb), are highlighted. The internal dipole points from the P700 center to the iron-sulfur clusters. Figure from [10]. (b): Energy level diagram for electron transfer in PSI upon excita on of the P700 reac on center. Figure from [9].

Recently, researchers have found indica ons that PSI is spin selec ve (Naaman et al.): It seems that the charge transfer prefers or polarizes electrons to have spins parallel to their momentum [9]. Proteins in the PSI complex have a α-helical secondary structure. It is thought that this chi- rality plays a major role in the spin polariza on, which is in accordance to observa ons in dou- ble stranded DNA [11]. However, there is some debate on the reliability of the results on spin- selec vity in electron transfer in PSI.

This research aims at implemen ng PSI in graphene devices for new photovoltaic systems and probing its charge transfer characteris cs. The ul mate goal would be to measure the spin selec- vity in an unambiguous way, which is described conceptually in sec on 2.3, and specifically in 2.3.2. First, a thorough analysis on the paper of Naaman et al. is made in order to understand whether their measured signal could really be from PSI. This is done in sec on 2.1 and 2.2. The experimental chapters 3 and 4 focus implemen ng PSI in graphene devices. In chapter 3 experiments on immobilizing and characterizing PSI monolayers on graphene are discussed. In chapter 4 the integra on of PSI in a graphene field-effect transistor (FET) is introduced.

1Debeye units: 1 D = 3.336· 10³⁰Cm = 2.082· 10⁻²e· nm

(7)

Chapter 2 Theory

2.1 Observa ons of spin-selec vity in PSI

Spin selec vity in electron transfer in PSI was first reported in 2014 by Naaman et al. [9]. The device structure used to observe it is depicted in figure 2.1a. The PSI complexes are oriented on a silver strip by means of short linker molecules. It lays across a grounded ferromagne c nickel strip, separated by an aluminum oxide tunnel barrier. The poten al difference between the silver and nickel is measured.

(a) (b)

Figure 2.1: (a): Schema c of the devices used to measure spin-selec vity in PSI. Upon illumina on the P700 reac on center is excited and an electron is transferred upwards to the iron-sulfur clusters. The remaining hole is then ﬁlled up by an electron from the silver below as described by the authors. (b): Measured voltage between the silver and nickel layer. If the nickel is magne zed up (blue-arrows), the voltage drops. This indicates that electrons with magne c dipole upwards are more likely to tunnel through the aluminum oxide layer, from the nickel strip into the silver, so there should be more empty spin-up states in the silver. Figures from [9]

Upon illumina on with a 660 nm laser, electrons are excited and move from the P700 reac on center (bo om) to the iron-sulfur clusters (top) in the PSI. The hole that is le in the P700 center is then filled by an electron from the silver strip. At this point, the silver resides at a higher poten al compared to the grounded nickel. An applied magne c field sets the magne za on of the nickel ferromagnet. The direc on of this polariza on is indicated by the blue arrows in figure 2.1b. As observed, a downwards magne za on yields a high poten al whereas an upward magne za on

(8)

2.1. OBSERVATIONS OF SPIN-SELECTIVITY IN PSI

results in a low poten al between the silver and the nickel. Thus, there is more tunneling from the nickel to the silver strip for the upward polariza on than for the downward one. This means that the electrons replaced in the silver have their intrinsic magne c dipole upwards, which should be because the PSI selects mainly this speciﬁc spin from the silver. Therefore, it is concluded that electron transfer in PSI is spin-selec ve.

This paper raises some ques ons. First of all, the method that was used to observe spin-selec vity is unconven onal [12]. This is a type of local spin valve geometry, where the spin-selec vity is observed by measuring the voltage across the electron transfer path. A be er way to measure the signal would be to build a non-local spin valve, where the signal is measured away from the actual spin injector contacts [13]. Pure spin current can be observed in such a device. The nonlocal spin valve geometry will be introduced in sec on 2.3.1, being one of the long-term goal of this research.

Another issue arises from the magne za on of the nickel strip. Due to its easy axis being in-plane, it cannot have a remnant magne za on out-of-plane [14]. Therefore, in order to polarize the nickel perpendicularly, an external magne c field needs to be applied constantly. This might very well influence the measurements. Measurements have been performed for the other orienta on of PSI as well. Here, the modula on of the poten al with magne c field direc on has the same sign and similar magnitude in both cases (fig. C.1b in the appendix). The paper ascribes this to the spin being selected parallel to the transfer direc on, but it might also be a mere artefact of the applied magne c field.

A third point is the nonzero oﬀset voltage for the modula on with magne za on in ﬁgure 2.1b.

Transferring an electron towards the top of the PSI complex would indeed give a poten al diﬀer- ence between the silver and the nickel. However, the electron at the top is not transferred further.

Therefore, the voltage should drop to zero as all the electrons transferred from the silver to the PSI eventually get replaced by electrons tunneling from the nickel strip. Thus, the signal should die out a er some me, but this is not the case. A possible explana on for this is that there is some non spin-polarized relaxa on path through the PSI itself, which makes it act like a spin current source toward the silver. This does not yet explain why there is a nonzero oﬀset, which indicates a misalignment in Fermi level between silver and nickel. There are some possible explana ons for this: it could, for instance, be a thermoelectric eﬀect from an in-plane temperature gradient.

For these reasons, our experiments aim to repeat the measurement of spin-selec vity in electron transfer in PSI using a non-local geometry. The contacts should exhibit perpendicular magne c anisotropy to have out-of-plane magne za on [15, 16]. The details of the actual device will be discussed in sec on 2.3.1. The next sec on will provide a thorough analysis of the spin-transfer dynamics in the device of ﬁgure 2.1a by means of a circuit model. This way, the order of magnitude of the spin current is reviewed to see if it is possible at all in PSI, and to see if the signal is large enough to be measured by a diﬀerent geometry.

(9)

2.2. CIRUIT MODEL

2.2 Ciruit model

In this sec on the spin current generated by PSI is es mated by use of a circuit model shown in (ﬁg. 2.3). As explained in the previous sec on, PSI extracts spin-up electrons from silver and due to internal relaxa on events they end up back in the silver, but they may have lost their spin orienta on. This gives a spin imbalance in the silver, depicted on the le in ﬁgure 2.2 [17].

Figure 2.2: Normal metal (NM, le ) in which a spin-imbalance has been created. When placed close to a ferromagnet (FM, right) polarized in the same direc on as the spin-imbalance, those spins in NM feel a poten al difference as they reside above the Fermi level µ0. This makes tunneling possible and fills up the ferromagnet above its Fermi level. Thus, a poten al difference of Vscan be measured between NM and FM. Figure from [17].

The quan es µ_↑, µ_↓, and µ0are the spin-up chemical poten al, the spin-down chemical poten al and the Fermi level of the system respec vely. Spin-up electrons in the silver reside at a poten al energy ∆µ = µ_↑− µ↓higher than spin-down electrons there, and µ_↑− µ0higher than the Fermi level of the nickel ferromagnet - assuming there is no bias applied between these two metals.

This imbalance ∆µ_↑↓ is the driving force for the current through the circuit in ﬁgure 2.3, which is a so-called spin current. If we split a charge-current I in a spin-up current I_↑and a spin-down current I_↓. We get

I = I_↑+ I_↓ and Is = I_↑− I↓ (2.1) where Is is the net spin current. In the limit that I = 0 and Is ̸= 0 we speak of a pure spin current. In our model in ﬁgure 2.3 the spin currents moves clockwise. An alterna ve view is that the spin-up current moves along the top (red) channel and the spin-down current moves along the lower (blue) channel.

Star ng in the PSI complex, a spin current Is(or spin-up charge current) is generated by absorp on of photons. It moves towards the contact with silver, modeled as contact resistance RcPSI↑, during which it might relax within the PSI itself through several channels. These are binned in the spin-ﬂip resistance Ri-sf. Next, the spins face resistance R_Ag↑of the silver layer, wherein they might also ﬂip.

We model this as spin-ﬂip resistance RAg-sfbetween the spin-up and spin-down path. Therea er, they go through the aluminum oxide tunnel barrier into the nickel. The tunnel resistance Rtb↑

depends on the magne za on of the nickel, being high if it is downwards (an -parallel to the electron spin) and lower if it is upwards (parallel to the electron spin). Finally, the spins end up in

(10)

2.2. CIRUIT MODEL

Figure 2.3: Circuit model based on the device from Naaman et al. (ﬁg. 2.1a). Upon illumina on, a spin current Ispropor onal to the number of absorbed photons per second will be generated within photosystem I. In order to neglect internal relaxa on paths, the PSI is considered to be a spin current injector with I_s^′ the net spin current injected into silver. Spin-ﬂip is allowed within silver via a channel modeled by RAg-sf. The voltage between the silver and nickel is measured in V .

the gold contact with resistance R_cNi↑, connec ng the nickel to the voltmeter as well as the ground.

Now, the spins follow the inverse path with its respec ve resistances RcNi↓, Rtb↓, RAg↓and RcPSI↓. Alterna vely, one might think of this path as spin-down electrons moving from PSI to nickel. The silver is also connected to the voltmeter by a gold contact, which is located some 1 mm away from the tunnel junc on. Basically, all spin current will have relaxed to zero in this contact, since the spin relaxa on length is 150 nm in silver [18, 19]. Therefore, the contact resistance can be regarded as very high (R_cAg → ∞). It averages the voltage from spin-up and spin-down chemical poten als in the silver, since that is what the voltmeter measures.

Consider a spin current I_s^′ into the silver from the PSI - we neglect the internal spin-relaxa on events within the PSI for the moment - we get for the measured voltage

V = I_s^′ ·1 2

(R_↑− R↓) R_Ag-sf

R_↓+ R_↑+ R_Ag-sf (2.2)

where R_↑ = R_tb_↑+ R_cNi_↑ and vice-versa. Since the signal of the spin selec vity in PSI is found by looking at the voltage diﬀerence for both magne za on direc ons, the signal becomes

V_signal = V^(ap)− V^(p) (2.3)

with V^(p) the voltage for parallel magne za on and V^(ap) for an -parallel. Unfortunately, this also gives an addi onal label to R_↑ and R_↓, which might be confusing. This is necessary though, because the arrow merely indicates the spin path upon which the tunnel barrier resistor acts.

It tells nothing about the state of the tunnel barrier, i.e. parallel or an -parallel. Fortunately, by imposing symmetry to the spins in this tunnel junc on we can redeﬁne these resistances as

R^(p)_↑ = R^(ap)_↓ ≡ R^(p) and R^(p)_↓ = R_↑^(ap)≡ R^(ap) (2.4)

(11)

2.2. CIRUIT MODEL

and therefore

V_signal = I_s^′ ·

(R^(ap)− R^(p)) R_Ag-sf

R^(ap)+ R^(p)+ R_Ag-sf (2.5)

Now we have to ﬁll in all the values. From ﬁgure 2.1a we es mate the signal to be Vsignal≈ 50 nV.

The resistances in the tunnel junc on as well as the spin-ﬂip resistance in silver will be es mated in the sec ons below.

2.2.1 Resistances in the tunnel junc on

As explained, the relevant resistances in the Al2O3/Nijunc on are the resistance of the tunnel barrier Rtb and the contact resistance RcNi between the nickel strip and a gold contact. Start- ing with the former we are interested in R^(p)_tb and R^(ap)_tb . From literature it is found that for a Co/Al₂O₃/Ni₈₀Fe₂₀ junc on, the resistance in the an -parallel conﬁgura on is 5 · 10⁴ Ωµm² with a magnetoresistance-ra o of 10% at room temperature [20]. The thickness of the oxide layer used in this reference is 5− 15 Å, which is comparable to 5 Å in the device from ﬁgure 2.1a.

In our case, the interface area between the silver and nickel strip is 2 µm². Therefore, we would have R^(ap)_tb = 25 kΩand R^(p)_tb = 22.5 kΩ. Note that these values are for a magne za on along an easy axis in a ferromagnet-insulator-ferromagnet junc on. The system that we are describing has a magne za on along the hard axis in a metal-insulator-ferromagnet junc on, where the metal has a spin imbalance shown in ﬁgure 2.2. The observa ons of this tunnel magnetoresistance are performed in a local geometry [21], so it also includes a term for the contact resistance. Therefore, it should be safe to assume that the RcNiis included in the found values.

2.2.2 Spin-ﬂip resistance in silver

A por on of the spins injected in silver will ﬂip and are lost before they can reach the tunnel barrier.

This has been modeled as a spin-flip impedance RAg-sf between the spin-up and spin-down path of the circuit model (fig. 2.3). The driving force behind this spin relaxa on is the spin-imbalance in silver (fig. 2.2), which is given by the poten al difference VAg-sf = ^∆µ_e^↑↓ between the two channels.

The current through RAg-sfequals

I_Ag-sf = e 2

∆n_↑↓

τ_Ag-sf (2.6)

with τAg-sfthe spin-ﬂip me in silver and ∆n_↑↓the number of unpaired spins. By introducing the mean energy spacing between non-degenerate states in silver ∆m= ^∆µ_∆n^↑↓

↑↓ we get [22, 23]

I_Ag-sf = e∆µ_↑↓

2∆_mτ_Ag-sf (2.7)

which then gives for the spin-ﬂip resistance R_Ag-sf = V_Ag-sf

I_Ag-sf = 2∆_mτ_Ag-sf

e² (2.8)

(12)

2.2. CIRUIT MODEL

Figure 2.4: Known electron transfer rates between the P700 reac on center and the iron-sulfur clusters in PSI. Figure from [26].

The spin-ﬂip me τAg-sfis 3− 7 ps as observed in mesoscopic silver [18, 24], we take τAg-sf = 7 ps.

A good approxima on of the mean energy level spacing ∆mwould be the inverse density of the states (_dn

dE

)₋₁

at the Fermi level in silver. Therefore, considering the free electron model we get

∆_m ≈ 2π² V

( ¯h² 2m_e

)³₂

√1

E_f (2.9)

with m the electron mass, Ef the Fermi level and V the volume of the part of silver right above the tunnel barrier. A jus ﬁca on for the appropriate value of V is given in appendix A. The spin- ﬂip resistance can now be calculated. We take V = 1 µm× 2 µm × 50 nm and Ef = 5.48 eV [25], so we get for the resistance RAg-sf ≈ 55 mΩ.

2.2.3 Es ma on of spin injec on current

Using the values found in the previous two sec ons, the spin injec on current from the PSI mono- layer into the graphene can be es mated: I_s^′ ≈ 17 µA. If we assume that there is one PSI monomer for every 100 nm², then each should produce a current of 0.86 nA. This should be com- parable to the current from photo-excita on and relaxa on in PSI. This current is propor onal to the excita on and relaxa on rates. We get these rates from ﬁgure 2.4. Relaxa on is dominated by the recombina on me τrel = 10 ms, which gives current Irel= 1.6· 10⁻¹⁷A. This makes the value received from the circuit model seem dispropor onately large.

It might be possible that the relaxa on rate is much higher due to the lack of top electron transport layer on the PSI monolayer. Therefore, we es mate another relaxa on current based on the total excita on and transfer me from P700 to the iron sulfur clusters τexcite-transfer ≈ 600 ns, we get Iexcite-transfer ≈ 0.27 pA. Unfortunately, this value is s ll several orders of magnitude too small

(13)

2.3. GRAPHENE

to match I_s^′ from the circuit model. As a final sugges on, it could also be that the iron-sulfur clusters are saturated with electrons a er the first couple of excita ons. The transfer towards those becomes impossible, thus excita on-relaxa on me can become on the order of tens of picoseconds, e.g. τexcite-relax ≈ 30 ps. This gives a current Iexcite-relax ≈ 5.3 nA, which is a bit higher than I_s^′. This is reasonable, since the spin-polariza on of the current generated by PSI does not have to be 100%. Reviewing this circuit model will give be er understanding in the validity of our es ma on. For now, the analysis with our circuit model indicates that spin selec vity in electron transfer in photosystem I would not be measurable with the device from Naaman et al. in figure 2.1a.

2.3 Graphene

In order to study the spin selec vity in electron transfer in PSI we turn to graphene for its large spin relaxa on length (2 µm at room temperature [27]). A conceptual device for these measurements based on a nonlocal spin valve geometry is introduced ahead in this sec on. Graphene is a single layer of sp2 hybridized carbon atoms, forming a hexagonal la ce structure. The first graphene devices were fabricated and characterized in 2004 by Geim, Novoselov et al. [28], who exfoliated graphene using their now famous ’Scotch tape’ method. The band structure of graphene consists of so-called Dirac cones at high symmetry points (K and K^′) in the first Brillouin zone [29], these are shown in the insets in figure 2.5a. Graphene has no band gap as the valence and conduc on band meet at a single point in every Dirac cone. At equilibrium the Fermi level of graphene is in this point, and therefore there are virtually no charge carriers present in this case; no holes because the valence band is completely filled; no electrons since the conduc on band is empty.

It is called the charge neutrality point for this reason.

(a) (b)

Figure 2.5: (a): Backgate dependence of the resistance in graphene. A nega ve voltage lowers the Fermi level below the charge neutrality point (le inset), this is the hole conduc on regime. Alterna vely, a posi ve backgate voltage raises the Fermi level (right inset) resul ng in electrons as charge carriers. Figure altered from [30]. (b): Schema c of a graphene ﬁeld eﬀect transistor with silicon dioxide as backgate dielectric and doped silicon as backgate.

For graphene devices with a typical field-effect transistor geometry (fig. 2.5b), the Fermi level can be tuned by adjus ng the backgate voltage. This is depicted in figure 2.5a. The resistance

(14)

2.3. GRAPHENE

of the graphene channel has a peak around zero voltage. This is the charge neutrality point, but it is also called a Dirac peak. The conductance is low in this point, because there are not many charges to move around. Thermal broadening and the uncertainty principle always give room to a few charge carriers, such that the resistance remains rela vely low compared to semi-conductor FETs in their off-state [31]. A nega ve backgate voltage will lower the Fermi level, resul ng in holes as free charge carriers in the graphene. Similarly, a posi ve voltage raises the Fermi level, resul ng in electrons as charge carriers. Like in semiconductors, the former case is called posi ve- or p-type and the la er nega ve- or n-type. Due to the electric field effect from impuri es in the backgate dielectric or junk on the graphene, such as polymers, the neutrality point will o en be shi ed from zero backgate voltage [32]. This effec vely results in p-doped (shi to right) or n- doped (shi to le ) graphene. The mobility of graphene is propor onal to the slope of the Dirac curve, reaching values of 125000 cm²V⁻¹s⁻¹ at room temperature [33]. Thus, the mobility of graphene as well as the free charge carrier type can be tuned by the backgate electrode.

2.3.1 Non-local spin valve

The observa ons of spin-selec vity in photosystem I, discussed in sec on 2.1 have been performed by local measurements [9]. This means that the signal is measured at the injector contacts, in which case the voltage drop from any charge current I is also measured rather than from spin current Is alone. In general, measuring the pure diﬀusive spin current away from the injec on contacts is accepted to be a robust and unambiguous way of observing a pure spin current [13].

The device for performing these measurements is called a nonlocal spin valve, which is shown in figure 2.6a. It consists of four (cobalt) ferromagne c contacts on graphene separated by a tunnel barrier (aluminum oxide). By varying the size of the ferromagne c contacts their coercive fields can be tuned, such that they switch magne za on direc on at different applied magne c fields. This means that the remnant magne za on of the contacts can be set independent of each other. If they all have parallel magne za on, the current source injects spin-up electrons¹in the graphene below FM3 and extracts spin-up electrons from FM4; effec vely injec ng spin-down electrons at FM4. The corresponding spin-up and spin-down chemical poten als, respec vely µ_↑ and µ_↓, are shown in figure 2.6b. At FM3 we have µ_↑ above the Fermi level µ0 and µ_↑ at FM4 lower than µ0. Consider the spin chemical poten al

µ_s = µ_↑− µ_↓ (2.10)

at FM3. It is nonzero, so it will diffuse to the le two contacts. A voltage difference from the drop of spin-up chemical poten al between FM2 and FM1 can be measured across these contacts, because they are magne zed up, and thus both probe µ_↑. If an applied magne c field now flips the magne za on of FM2 and FM4, we get the situa on depicted in figure 2.6c. Now FM2 probes µ_↓, which is lower than µ0. This means that the voltage between FM2 and FM1 must flip sign. This indeed turns out to be the case [27], as can be seen in figure 2.7.

1The arrow indicates the direc on of the electrons. Of course, the charge current is the other way around.

(15)

2.3. GRAPHENE

(a) Nonlocal spin valve geometry

(b) Conﬁgura on 1

(c) Conﬁgura on 2

Figure 2.6: (a): Nonlocal spin valve geometry. It consists of a graphene sheet on an insula ng substrate with four ferromagne c contacts (FM1 through FM4), separated by an aluminum oxide tunnel barrier. (b): Chemical poten als for all ferromagnets magne zed upwards. Spin-up is injected in FM3, raising µ_↑. This imbalance diffuses outwards to FM1 and FM2, it may be measured as voltage difference between them. (c): Chemical poten als for a another configura on. FM2 and FM4 are an -parallel to FM1 and FM3. FM2 now measures the poten al µ_↓, so the voltage between FM2 and FM1 flips sign. Figures from [34].

Diffusion of the spins goes according to the Bloch equa on [34]. At steady state with no applied magne c field it reduces to an ordinary diffusion equa on:

D∇²µs= µ_s

τ (2.11)

with D the diffusion coefficient and τ the spin relaxa on me. Since we are mostly interested in the diffusion to the le , we need the one-dimensional solu on to this problem. We get

Dd²µ_s dx² = µ_s

τ ⇒ µs(x) = µ_s(0)· e^−x/^√^Dτ (2.12) with√

Dτ = λthe spin relaxa on length. The value for this length is quite long in graphene, as

(16)

2.3. GRAPHENE

Figure 2.7: Nonlocal signal (defined in eq. 2.8) during a magne c field sweep. To the right (green) the ferromagne c contacts are switched one-by-one, because they have different coercive fields, a large peak is observed when FM2 is an -parallel to the other contacts. To the le (red) again a large peak is measured when FM2 is an -parallel to the rest. Figure from [27].

men oned 2 µm and possibly even longer. This is quite far compared to other metals like silver, which has a relaxa on length of only 150− 270 nm at room temperature [18, 19, 24]. This makes graphene the ideal material for detec ng a spin-injec on current from PSI.

2.3.2 Device for measuring spin-selec vity

A goal will be to build a nonlocal spin valve with photosystem I. The device geometry is similar to the one in figure 2.6a. On the right side, instead of a current source connected to two ferromagne c contacts, PSI will inject charge carriers into graphene. If these are spin polarized, a spin chemical poten al (eq. 2.10) will accumulate in the graphene below the PSI. This imbalance will diffuse to the two ferromagne c contacts, which are separated from the graphene by an insula ng layer. By switching the polariza on of the ferromagnets from parallel to an -parallel with respect to each other (le two contacts fig. in 2.6b and 2.6c), the spin-signal can be measured.

According to literature the spins in the current from PSI are oriented along the transfer direc on [9]. This means that they are poin ng out-of-plane, along the ver cal axis in figure 2.6a. The easy axis for most ferromagne c contacts, like cobalt and nickel, is in-plane. This means that they only have a remnant magne za on for in-plane polariza on direc ons, but in order to magne ze the contact ver cally an applied magne c field must always be present. This is something we wish to avoid, because it introduces a precession term in equa on 2.11, making the results harder to interpret. More importantly, it might introduce some unknown effects within PSI, which can influence the measurement. The solu on to this problem will be to introduce ferromagne c contacts with perpendicular magne c anisotropy. These are more complicated structures of e.g. Co/Pt mul layers [15, 16, 35, 36].

Another challenge is to generate enough charge transfer to actually build up a measurable spin-

(17)

2.4. RESEARCH GOALS

chemical poten al in the graphene. Basically, this means building a small photovoltaic cell onto the graphene with PSI as ac ve layer [7]. In order to achieve this, all PSI systems on graphene must be oriented in the same direc on, because otherwise the transferred charges will cancel themselves. Next, a top-layer is needed to have con nuous electron transfer, extrac ng the electrons on one side and ﬁlling the holes on the other. This top layer should be transparent to be able to illuminate the PSI. Therefore, it would be appropriate to also use graphene as top-layer.

This means transferring a sheet of graphene onto a PSI monolayer [37].

Another important point is to analyze what happens to the graphene once PSI is immobilized on it. Does it s ll act as graphene or not? It could be that PSI or linker molecules covalently bond to the graphene sheets. This would locally re-hybridize the carbon atoms, which likely influences the electrical proper es of graphene. However, monolayers of PSI contain roughly one trimer complex per 10⁻²µm². This means that there are some 10⁵−10⁶carbon atoms, as graphene has bond length 1.42 Å per PSI system [38, 39]. Therefore, the direct effect on the graphene sheets will be limited. However, it might s ll have some doping effect on graphene.

Based on the es ma on of spin-injec on from the circuit model (sec on 2.2.3) we can es mate the possible spin signal measured by this device to see whether the experiment is feasible or not.

Previous research has shown that the nonlocal graphene spin valves from sec on 2.3.1 have a signal ∆RN L = 5 Ω[27]. When considering ﬁgure 2.6a, this nonlocal resistance RN Lis deﬁned as

R_{N L}= VF M 2− VF M 1

I (2.13)

In these measurements the relevant spacing, between FM2 and FM3, was 3 µm. By taking the in- jec on current I = I_s^′ ≈ 17 µA from sec on 2.2.3, we get for the measured poten al diﬀerence V_{F M 2}− VF M 1 ≈ 85 µV. This is deﬁnitely a measurable signal. The ferromagne c contacts FM3 and FM4 are now replaced by the PSI monolayer, bound with pep des. Note that PSI should inject a highly spin-polarized current, performing be er than the ferromagnet/tunnel barrier contacts.

However, the pep de binders might induce some relaxa on. Applying a top-layer of graphene to the PSI would improve the injec on current. Placing the FM2 contact closer to the PSI monolayer should improve the measurable voltage. In conclusion, the spin injec on by PSI should be measurable with this device if the es ma on I = I_s^′ ≈ 17 µA (sec on 2.2.3) is correct.

2.4 Research goals

As men oned, fabrica ng and characterizing a device to measure the spin selec vity in PSI is the ul mate goal of this research topic. Combining PSI with graphene without a liquid environment is quite a new direc on of study. Therefore, this report is limited to several subtopics comprising the final goal. First, the PSI monolayer needs to be self-assembled onto graphene, ideally with the same orienta on among every PSI trimer complex. The method and results on these experiments will be discussed in chapter 3. Next, the effect of PSI on the electrical proper es of graphene will be analyzed by means of a field-effect transistor in chapter 4.

(18)

Chapter 3 Self-assembled Monolayer of PSI on Graphene

Studies on photosystem I monolayers have been performed on several substrates already. For instance, bio-photovoltaic cells incorpora ng PSI have been researched [7], as well as the charge transfer characteris cs in this photosystem [9]. Devices from these monolayers on graphene have not yet been studied though. Layers of PSI on graphene and its charge dynamics have been fabricated, but only in the presence of a mediator solu on [40]. This chapter will show results from self-assembly of PSI monolayers on chemical vapor deposited (CVD) graphene for fabrica on of dry devices. We aim to have all PSI complexes aligned on the graphene, meaning that their intrinsic electric dipoles (ﬁg. 1.2a) point in the same direc on. In this way charge transfer will be op mal for applica ons like photovoltaics and spintronics. Tunneling Atomic Force Microscopy will be used to analyze this orienta on.

3.1 Orien ng PSI on graphene

Previous studies used short linker molecules to orient PSI on substrates [9]. We use pep des, short chains of amino acids, to bind the PSI to graphene. The wide variety in pep des allow us to select those that are best fit to our situa on. On one hand, we need to bind to graphene, on the other to the iron-sulfur clusters in PSI in order to orient the dipole. With these criteria in mind, two pep de chains are selected and bound together by a linker¹. The PSI itself is extracted from T. Elongatus cyanobacteria [42]. The photosystem I complex from these bacteria has been well described [43]. It exists mainly as a trimeric complex with height 6 nm and diameter 25 nm [44], shown schema cally in figure 3.1. A er extrac on, the PSI placed in a buffer solu on², which aims to mimic the condi ons within the thylakoid membrane of the cyanobacteria.

3.1.1 Self-assembly process

Both the pep de and PSI trimer complex solu ons are combined in a one-to-one ra o and diluted with deionized water to reach a concentra on of 0.1 M. The solu on is then shaken for one hour at a temperature of 5. The product of this shall be referred to as the PSI-pep de solu on in the remainder of the text.

1Amino acid chain: GAMHLPWHMGTL-G-DPALHETKGAQI [41].

220 mmolHEPES, 10 mmol MgCl2, 10 mmol CaCl2, 500 mmol mannitol, pH 7.5 [45].

(19)

3.2. ATOMIC FORCE MICROSCOPY

Figure 3.1: Photosystem I trimer. It has thickness 6 nm and diameter 25 nm. Figure from [46].

Now, the solu on is brought in contact with graphene to start the self-assembly process. This has been done in several ways. For small enough substrates, like graphene on copper foils, they are submerged into vials containing the PSI-pep de solu on. For larger substrates, such as several pieces of Si/SiO2/graphene, a drop is deposited on the graphene. The dura on of the process is the same in both cases: the drop is removed a er 2 hours. Next, the samples are dried with nitrogen ﬂow and the self-assembly is complete. Aside from PSI-pep de, also pep de-only and PSI-only solu ons are applied onto graphene for reference experiments.

3.2 Atomic Force Microscopy

The monolayers of PSI on graphene on silicon dioxide substrates have been imaged using tapping mode Atomic Force Microscopy (AFM). The height proﬁle on the substrate is measured via the deﬂec on of an oscilla ng can lever, which scans across the substrate. Depending on the quality of the p below the can lever, nanometer resolu ons can be achieved [47].

3.2.1 Results

The AFM images from four different samples, i.e. PSI-pep de, PSI-only, pep de-only, and pris ne graphene, are shown in figure 3.2. The monolayers in figure 3.2a and 3.2b consist of li le spheres with diameters ranging between 25 and 50 nm, which indicates that these are PSI trimers. Unfor- tunately, the spheres show up slightly too large. This might be some artefact from the tapping mode AFM or from aggrega ons of mul ple PSI complexes and pep des. This shall be discussed in sec on 3.4. Although the coverage of PSI varies throughout different areas on the same substrate, it can be seen that there is more PSI present for the PSI-pep de (fig. 3.2a) sample than for the PSI-only (fig. 3.2b) sample. This should be due to pep des binding the PSI to the graphene in the former case, whereas the PSI is fixed to graphene by Van der Waals’ interac ons in the la er. The pep de-only sample (fig. 3.2c) lacks the presence of the PSI spheres. This is another

(20)

3.3. TUNNELING AFM

indica on that we have a PSI layer in ﬁgure 3.2a, since we now have no PSI in the self-assembly solu on. There are some ny specks of about 10 nm visible, which may point to coiled up pep de chains, since they are about 2 nm long. Moreover, the features of CVD graphene become visible in this image and also in the pris ne graphene one (ﬁg. 3.2d), because polymer residues show up as thick lines on the graphene surface [48].

(a) PSI-pep de (b) PSI-only

(c) Pep de-only (d) Pris ne graphene

Figure 3.2: AFM height profiles of (a) PSI-pep de, (b) PSI-only, (c) pep de-only monolayers on CVD graphene/silicon dioxide substrates. Profile (d) is from a pris ne graphene sample. Self-assembly occurred inside a droplet on the surface. The insets in (a) and (b) are zoomed-in images of the same profiles.

3.3 Tunneling AFM

In the previous sec on, the height profile of the monolayers on graphene have confirmed the presence of PSI. Now the orienta on needs to be confirmed, which is done by Tunneling AFM (TUNA). This is a type of conduc ve probe AFM (CP-AFM) where a voltage bias can be applied

(21)

between the p and the substrate, upon which a current can be measured. The setup is depicted schema cally in ﬁgure 3.3a.

(a) (b)

Figure 3.3: (a): Schema c of the Tunneling AFM setup with our sample. A voltage bias can be applied between the p and the gold contact upon which a current can be measured. The aim is to orient all PSI systems to have their dipole downwards, thus P700 on top and the iron-sulfur clusters, [4Fe-4S], at the bo om. (b): A typical I-V curve obtained by this measurement, an asymmetry can be seen when comparing the current magnitude at posi ve and nega ve bias (e.g.±0.5 V).

In our sample the photosystem I is self-assembled onto a graphene layer on a silicon dioxide substrate. A 50 nm gold contact is evaporated on the graphene with a 5 nm tanium adhesion layer.

Silver paste is used to connect this contact to the sample holder for electrical contact. A voltage bias can now be applied between the CP-AFM p and the graphene. We hypothesize that this will induce a tunneling current through the PSI [49]. Because of the internal dipole in the complex, there will be an asymmetry in the conductance of the PSI. A reverse bias should generate a larger current than a forward bias if the PSI orienta on is like the one in ﬁgure 3.3a, which is what we want to achieve. From this asymmetry in I-V characteris cs the orienta on may be measured.

Measurements have been performed as follows: First, an ini al height scan was performed in order to conﬁrm the presence of PSI. The limited placement precision of the AFM p combined by some movement of the PSI systems make touching upon a single PSI trimer unreliable. Therefore, the AFM p is touched upon a grid of points within the ini al scan region. This point-and-shoot method should make sure that we touch upon a PSI complex every once in a while. The presence of a PSI system between the p and the graphene can be seen from the magnitude of the current.

For instance, a high current with a rela vely linear I-V curve indicates to a short, where the p probably touches directly onto the graphene. A lower current with a nonlinear I-V-curve, on the other hand, points to a tunneling mechanism, which should mean that the p is touching a PSI system.

The asymmetry of the I-V-characteris c can be observed via the rec ﬁca on valueℜ [49]. It is deﬁned as follows

(22)

ℜ = I₋

I₊ (3.1)

in which I+and I₋are measured values for the current at some posi ve voltage and its nega ve counterpart, respec vely. In our case, we take the currents at±0.5 V. We assume that contact asymmetry plays a minor role in this setup. Thus, measuring a rec ﬁca onℜ > 1 means that the PSI dipole would be oriented downwards, whereas a rec ﬁca onℜ < 1 means that it would point upwards. For this reason, it is best to consider a logarithmic scale logℜ.

3.3.1 Results

A er collec ng thousands of I-V curves, the rec ﬁca on values can be summarized in histograms.

However, before doing so some filtering needs to be applied, because many of the measured I-V- curves show breaks (fig. 3.4a). In this situa on the AFM p makes no contact with the sample at all. In addi on, many curves show shorts (fig. 3.4b), in which case the p contacts the graphene directly and the current becomes higher than the compliance limit of the AFM setup. This can be seen as a satura on in the I-V curve, but it has no physical meaning compared to the applied bias voltage. On average, 41% breaks 39% shorts are filtered out.

(a) Break (b) Short

Figure 3.4: Filtered curves for Tunneling AFM measurements. (a): Break. The CP-AFM p is somehow obstructed from touching anywhere on the sample, so no current can be measured while sweeping the bias voltage. The oﬀset from zero current comes from the resolu on of the setup. (b): Short. The p is touching directly upon the graphene having resistance of kΩ order. The current exceeds the sensi vity limit, I < 10⁻⁶A, which acts as a compliance limit.

The current values at all swept voltages are recorded, but they have no physical meaning regarding the experiment.

The histograms of the R-values a er filtering are shown in figure 3.5 for four different samples:

one with a PSI-pep de monolayer, one with PSI-only, one with pep de-only, and a pris ne graphene sample. The magnitude of the current log|I−| at which these ℜ-values occur is given along the ver cal axes of ﬁgures 3.5c and 3.5d. It can be readily seen that the logℜ values are mostly above

(23)

3.4. DISCUSSION

zero and around 0.2 for the PSI-pep de sample (ﬁg. 3.5c), which indicates that the PSI is oriented on the graphene for that situa on. This is not the case for the other three situa ons, where the logℜ values are sca ered across a wider range. The yield of valid data points is by far highest in the PSI-pep de situa on, which is due to the fact that the coverage of PSI is highest with pep de binders, as was shown in the previous sec on (ﬁg. 3.2).

(a) PSI-pep de (b) PSI-only

(c) PSI-pep de (d) PSI-only

Figure 3.5: (a) & (b): Histograms of the rec ﬁca on values for PSI-pep de and PSI-only samples respec vely. c)

&d): 2D-histograms of both the rec ﬁca on value and the order of magnitude of the current for PSI-pep de and PSI-only samples respec vely. It can be seen that the current for PSI-pep de samples is consistently several orders of magnitude larger than the PSI-only samples.

3.4 Discussion

In this sec on the interpreta on of the data will be discussed. The presence and coverage of PSI on our graphene layer as observed from the height proﬁle shall be discussed ﬁrst. The following sec on will address the observa ons regarding the orienta on of the PSI monolayer.

(24)

3.4. DISCUSSION

3.4.1 Size and coverage PSI monolayer

The measured size of the PSI spheres in sec on 3.2.1 is on the high end. Literature states that it should be 25 nm [44], but we measured sizes anywhere between 25 nm and 50 nm. One possible explana on for this is that tapping mode AFM structurally overes mates the in-plane dimensions of an object. This comes from the fact that when the p passes over an object it first goes upwards following the shape nicely. Then, right before moving downwards, it has a li le overshoot as it flies across the edge of the object. Making the p press harder by reducing the amplitude setpoint and decreasing the scan speed goes a long way in reducing this overshoot. However, it seems that in the measurements from figure 3.2 it was not fully eliminated. In addi on, p convolu on plays a role. All height profiles from tapping mode AFM measurements are basically a convolu on of the p and the substrate. This leads to an overes ma on of horizontal dimensions. Measurements would be accurate for an infinitely sharp p, as it would act like a Dirac delta func on in the convolu on. Aggrega on of PSI trimers might also play a role here; it could be that some mes a couple of PSI complexes cluster together.

The coverage of PSI varies across the graphene surface. Areas that seem to have a color more towards green in the op cal microscopy image are more densely packed with PSI systems than other areas (fig. C.4 in appendix). It is likely that this nonuniform coverage is due to the fact that self-assembly occurred in a drop on the substrate. In order to keep as much of the solu on on the surface as possible, a droplet of about 60 µL was applied to the sample. Most of the mes it remained intact on the surface. However, some mes it flowed off, leaving only a film of solu on on the substrate. Either samples were used regardless. Though, it might be that such a film of PSI solu on results in a much more uniform layer of PSI on graphene.

PSI also ends up on the gold contacts during the self-assembly process. This is due to the large electron aﬃnity of gold [50]. Moreover, in later samples it turned out that PSI can also be found on silicon dioxide, but usually not everywhere. With all these diﬀerent regions it becomes hard to dis nguish the graphene, the silicon dioxide, and the PSI covered silicon dioxide from each other.

For clearance, in ﬁgure C.4 in the appendix the diﬀerent regions are marked and discussed.

Some a empts have been made to deposit PSI on exfoliated graphene ﬂakes with area about 5 µm²: once on graphene silicon dioxide and once on gold. Unfortunately, in both cases it was observed with AFM that there is no PSI present on graphene, some on SiO2and lots on gold. This seems come from the hydrophobic nature of graphene, expelling the PSI-pep de solu on [51].

This does not occur for the CVD graphene samples, because there is nowhere else to go for the liquid on such large sheets (1 mm²).

3.4.2 Orienta on PSI trimers

It remains hard to conﬁrm the orienta on of PSI due to the uncertainty having PSI between the AFM p and the graphene in this method. As explained in sec on 3.3, many points have been scanned and some ﬁltering is applied. The I-V curves vary a lot throughout the dataset. However,

(25)

3.5. CONCLUSION

even though this is the case, a clear diﬀerence in the spread of rec ﬁca onℜ can be seen when comparing PSI-only and PSI-pep de samples. The order current magnitude is 10⁻⁷ for the PSI- pep de sample and 10⁻⁹to 10⁻⁸ for PSI-only, which indicates a be er electrical contact in the former case. Assuming that electronic transport indeed occurs through a PSI trimer, its resistance would be on the order of 10 MΩ. Characterizing a pep de-only sample will be a good reference experiment to learn about the transport paths. Unfortunately, it has not yet been measured properly, because the current through those layers is too high for the sensi vity of the setup.

This can be adjusted, so a er tuning the setup, these measurements will be possible.

As explained, the PSI-only monolayer seems not to be oriented, having logℜ values of zero on average. However, this results is a bit strange for the PSI-only sample: It seems more logical that, rather than a distribu on of values around logℜ = 0, we should get two dis nct peaks; one for the up-oriented PSI trimers and one for the down-oriented ones. This might be explained by considering that the photosystem can also be oriented sideways. However, then it should show up with more depth on the AFM images, since the diameter of the PSI trimer is some four mes as large as its thickness. Such a clear difference could not be dis nguished from the height profiles (fig. 3.2).

Another point of improvement is the measure of asymmetry. Up to now we have used the rec- fica on value defined in eq. 3.1. This includes taking the current at two equal and opposite voltage values. However, this voltage is chosen rather arbitrarily at±0.5 V. The noise that may occur in this point is filtered by taking the average value of the current for five values around this voltage. Unfortunately, noise may also show up in the form of oscilla ons in the I-V curve (fig. C.2 in appendix), which s ll influences the results. A way to circumvent this problem is to integrate the current on the le and right half of the I-V plot, and dividing by each other. Another analysis would be to fit the curve to a func on based on tunneling currents, i.e.

I = f (V − δ) (3.2)

Now the asymmetry is given by the horizontal shi δ of this func on. This way, one considers the complete I-V curve in the analysis. Moreover, this measure of asymmetry δ is not inﬂuenced by the considered measurement domain, as is the case when integra ng the le and right side of the I-V curve.

3.5 Conclusion

We were able to self-assemble PSI onto sheets of CVD graphene using pep des. Conduc ve probe AFM measurements indicate that the trimers within the monolayer are oriented onto graphene.

However, measuring reference samples and improved analysis should lead to more conclusive observa ons. The next chapter will discuss the use of self-assembled monolayers of PSI on graphene in a ﬁeld eﬀect transistor device.

(26)

Chapter 4 Graphene Field-Eﬀect Transistor with Photosys- tem I

Immobilizing PSI on graphene might influence its electrical proper es as it has a built-in dipole of 1000D [8]. Therefore, we have inves gated the effect of PSI on graphene by means of graphene based Field Effect Transistor measurements [31]. As explained in sec on 2.3, the carrier density in graphene can be tuned by a ga ng electrode. The device layout for these measurements is depicted in figure 4.1. The large area channels of 1 mm²can only be achieved using CVD graphene.

These sizes are preferred, since it seems that the immobiliza on of PSI becomes increasingly harder upon using smaller graphene areas as discussed in 3.4.1.

4.1 Fabrica on

Star ng from commercially available CVD graphene on silicon dioxide¹substrates from Graphene Supermarket and Graphenea, several FETs have been fabricated. It has been shown in the past that the graphene from these companies is of sufficient quality for the ga ng experiments shown in figure 2.5a [52, 30]. The fabrica on process is depicted schema cally in figure 4.1b. Shadow masks are used to create the pa erns in graphene. The 10× 10 mm²graphene on silicon oxide is first etched in two steps to get four separate 1× 7 mm²strips of graphene (fig 4.1c) by Reac ve Ion Etching (RIE). This process was performed with oxygen plasma at a 9 sccm flow, 0.009 mbar pressure and 40 W power for the dura on of 20 seconds per step. Next, the substrates were annealed in a tube furnace for 10 hours at 250 C with a 20% Argon/Hydrogen flow. This procedure should remove most polymer residues, which may be present on the graphene [48]. During the next step contacts are created by consecu vely evapora ng tanium (5 nm) and gold (50 nm) on the graphene using a Temescal e-beam evapora on system (TFC2000). Now we have four strips with three FETs each. A er this the substrates were divided into two pieces along the red line in figure 4.1c to have twice as many samples for self-assembly.

The PSI was immobilized on the graphene using a procedure similar to the one explained in sec on 3.1.1: A drop of the PSI-pep de solu on was applied to the substrate, such that self-assembly will take place on the graphene channels. PSI-only as well as pep de-only solu ons are used to fabricate for reference. A er self-assembly the samples were glued onto a chip carrier using silver

1Oxide thickness is 285 nm and 300 nm for Graphene Supermarket and Graphenea respec vely

(27)

4.2. MEASURING THE DIRAC CURVE

(a)

(b)

(c)

Figure 4.1: (a): Side view of the CVD graphene transistor device geometry. The channels all have length and width 1 mm. (b): Fabrica on steps. First, the graphene is etched into four strips by plasma etching using two consecu ve shadow masks. Next, the tanium-gold contacts are evaporated onto the strips using the third mask. Figure from [30]. (c): Top view of the device substrate. A er fabrica on they are cut in two pieces along the red line to have 6 FET devices each.

paste, and wire bonding was done to have electrical contact between the gold contacts on the devices and the chip carrier pins.

4.2 Measuring the Dirac curve

In this sec on measurements are shown from the back-gate dependence of the sheet resistance.

These were performed at high vacuum. The channel resistance was measured using a lock-in am- pliﬁer setup; current was kept at 100 nA modulated at a frequency of 165.48 Hz unless stated oth- erwise. The backgate voltage was swept using a Keithley 2410 with a compliance limit of 100 nA and a maximum ramp of 0.5 V/s in series with a 1 MΩ resistor.

The results for four devices: Pris ne graphene, Pep de-only, PSI-only, and PSI-pep de, are plo ed ﬁgure 4.2. All are made from Graphene Supermarket substrates. Since we are dealing with square channels, the sheet resistance of graphene equals the channel resistance, as Rsheet = R_channel·^W_L with W and L respec vely the width and length of the channel.

The sheet resistance of the graphene is back-gate dependent. However, the Dirac point cannot be seen, as it is located too far to the right. This indicates a high p-doping of the graphene, as explained in chapter 2.3. Increasing the back-gate voltage past 80 V will likely lead to breakdown of the gate-dielectric (SiO2) [52], thus the peak is not measurable. It seems that the graphene is largely unaﬀected by the applied monolayers. Although, the curves are a bit steeper for the

(28)

4.2. MEASURING THE DIRAC CURVE

Figure 4.2: Backgate dependence of the sheet resistance in four diﬀerent samples: (a) pris ne graphene, (b) pep de- only (lock-in frequency 29.564 Hz, me constant τ = 300 ms), (c) PSI-only, (d) PSI-pep de. The blue and yellow curves are 2-probe (appendix B) measurements from the side channels A and C, the red curve is from a 4-probe measurement across the center channel B.

PSI-only and PSI-pep de samples in ﬁgure 4.2c-d, indica ng a shi to the le compared to ﬁgure 4.2a-b.

4.2.1 Earlier device

An earlier device made from a Graphenea substrate showed some interes ng results. The fab- rica on was similar to sec on 4.1, except this me the substrates were not annealed in a tube furnace. Rather, it was ﬁrst placed into a lock-in setup, pumped to high vacuum and annealed at 400 K in three steps, each about four hours. The backgate dependence was characterized: it was similar to ﬁgure 4.2a. Next, the device was removed from the setup. Now the PSI-pep de solu on was used to self-assemble a monolayer on the device without removing it from the chip carrier. A er two hours of self-assembly, the device was dried with pressurized air and loaded into the setup again.

The results are quite surprising: the Dirac point is now around 6 V for all four working devices, whereas before immobiliza on it was shi ed over 80 V to the right. The peaks are shown in ﬁgure 4.3. Some measurements were only possible on mul ple channels at once due to broken contacts.

All measurements are normalized with respect to channel length, since the sheet resistance is plo ed². A piece of aluminum foil covered the sample window of the setup, shielding it from

2The ’Coulomb’ measurement setup has a ﬁlter resistance of 1 kΩ for the lock-in probes. Therefore, also a cor- rec on of 2 kΩ is applied for these 2-probe experiments.

(29)

4.3. ILLUMINATION EXPERIMENTS

Figure 4.3: Backgate dependence of the sheet resistance in six diﬀerent channels on the same substrate a er self- assembly of a PSI-pep de monolayer. (a) and (e) comprise two channels, (b) is measured along three channels. The blue curves are obtained while keeping the substrate dark. The red curve was observed soon a er illumina ng the sample with ambient light. The yellow curve was captured some me later, s ll under illumina on. All measurements were by 2-probe conﬁgura on. The lock-in frequency was 282.9 Hz and me constant τ = 30 ms.

illumina on. In order to test the response of PSI on graphene to ambient light, it was removed and measurements were repeated. The blue and red curves in figure 4.3 represent these situa ons. It can be seen that the Dirac peak shi s to the right in all cases. However, the sample has not been measured in dark condi ons a er this. The observed shi might therefore also be a ributed to other effects, such as dri from trap states in the gate dielectric. Moreover, it could be seen with op cal microscopy that all channels on this substrate were flawed a er self-assembly, meaning that graphene was missing in some places (fig. C.5 in appendix), yet s ll a clear and con nuous conduc on path between the contacts could be dis nguished. The graphene on the devices from figure 4.2 was, however, s ll fully intact a er self-assembly. The next sec on will describe some be er defined illumina on experiments with those devices.

4.3 Illumina on experiments

PSI generates an electron hole pair upon excita on. This should change its dipole a bit. Thus, for monolayers of PSI on graphene, the poten al diﬀerence between PSI and graphene should change upon illumina on. The result will be a shi of the Dirac peak, since the doping level in the graphene is now diﬀerent. Although this peak could not be measured in the devices from sec on 4.2, the shi of the curve itself might s ll be visible.

Electron and Spin Transport in Graphene and Metallic Channels with Photosystem I Monolayers