Dynamics of Bloch oscillating transistor near bifurcation threshold Jayanta Sarkar,1 Antti Puska,1 Juha Hassel,2 and Pertti J. Hakonen1 3 1Low Temperature Laboratory, O.V.Lounasamaa Laboratory, 1 0 Aalto University, P.O. Box 13500, FI-00076 AALTO, Finland 2 n 2VTT Technical Research Centre of Finland, a J P.O. Box 1000, FI-02044 VTT, Finland 3 2 (Dated: January 24, 2013) ] l Abstract l a h Tendency to bifurcate can often be utilized to improve performance characteristics of amplifiers - s e or even to build detectors. Bloch oscillating transistor is such a device. Here we show that bistable m . behaviour can be approached by tuning the base current and that the critical value depends on t a m the Josephson coupling energy E of the device. We demonstrate record-large current gains for J - d device operation near the bifurcation point at small E . From our results for the current gains at J n o various E , wedeterminethebifurcation thresholdontheE -basecurrentplane. Thebifurcation J J c [ threshold curve can be understood using the interplay of inter- and intra-band tunneling events. 1 v 6 4 5 5 . 1 0 3 1 : v i X r a 1 In small Josephson junctions, charge and phase reveal their conjugate nature in sev- eral macroscopic phenomena. The quantum nature of the phase variable (ϕ) was shown in macroscopic tunneling experiments [1], while its conjugate relationship to the charge has been shown in many consequent studies [2]. One of the consequences of the charge-phase conjugate relationship is the Coulomb blockade of Cooper pairs which arises in ultra small Josephson junctions having a capacitance (C) in the femtoFarad range [3, 4]. Charging en- ergy and the Josephson coupling energy E (ϕ) = E cosϕ are the competing energy scales J J − associated with these two variables. Accordingly, the Hamiltonian for the small Josephson junction contains a periodic potential, and hence, Bloch states with band structure appear. These bands are analogous to the conduction electron energy states in solid state physics [5, 6]. The Bloch Oscillating transistor is a three-terminal mesoscopic device which is based on the dynamics of the Bloch bands in a voltage biased Josephson junction (JJ) in a resistive environment [7, 8]. The operation is due to an interplay of coherent Josephson phenomena and Coulombic blockade of charge transport which is controlled by single electron tunneling events. The device can be viewed as a charge converter of single electrons, induced from the base electrode, in to a sequence of N sequential Cooper pair tunneling events, i.e., Bloch oscillations on the emitter terminal with a Josephson junction. The current gain is ideally given by β = 2N + 1. The number of Bloch oscillations is limited by inter-band transitions caused by Landau-Zener (LZ) tunneling which depends exponentially on the band gap between the ground and excited states of the Josephson junction. This simple picture has been found to correspond quite well to the measured current gain [9]. Incoherent tunneling of Cooper pairs and electrons, however, complicates the basic BOT operation. The interaction of tunneling electrons or Cooper pairs with the electromagnetic environment, has been demonstrated to be strong in small tunnel junctions, both in the normal and superconducting states [10, 11]. Inelastic effects may, for example, limit the lifetime of the Coulomb blockaded state and, consequently, bias-induced changes in the inelastic tunneling rates can cause large modifications in the operating point, and thereby contribute to the current gain of the BOT. These effects, in fact, are the foundation for bifurcation in the BOT operation because they allow the existence of two steady states at a fixed base current I . The existence of a bifurcation point is important as, with proper B design, the vicinity of such a point can be employed to improve the characteristics of the 2 BOT. In this paper, we investigate experimentally the bifurcation threshold in the BOT, anddemonstrate recordlargecurrent gainsforsmall-E -device operationnear thethreshold. J From our results for the current gains at various E , we determine the bifurcation threshold J curve on the I – E plane. The measured transition curve can be qualitatively explained B J using a simple analytic approach, in which intra-band transitions are taken in to account phenomenologically, together with the transition rates due to inelastic tunneling. This paper is organized as follows. In Sec. I, we first outline the basic principles for understanding the electron tunneling dynamics in a Bloch oscillating transistor. We will concentrate to the dynamics near the bifurcation point at which the current gain of the device diverges. Our analytic model is verified using numerics with a similar approach as done in Refs. [8, 9, 12]. Sec. II will describe sample manufacture and experimental measurement techniques. Experimental results are presented in Sec. III. We will present data on the current gain at various values of Josephson energy, and construct a curve for bifurcation threshold on E vs base current plane. The relation of current gain with the J distance from the bifurcation point is also studied in detail. In Sec. IV, we discuss our results in the light of analytical and numerical calculations. I. THEORY A. Band model of mesoscopic Josephson junctions In mesoscopic tunnel junctions, the discreteness of charge starts to play a role via the Coulomb energy E = Q2, where C is the capacitance of the junction and Q is the charge on C 2C the capacitor plates. In quantum theory, charge is described by the operator Q = i2e ∂ , − ∂ϕ where ϕ denotes the phase difference of the order parameter fields across the junction. This b operator is canonically conjugate to ϕˆ, i.e., [Qˆ,ϕˆ] = i2e. Hence, there is a Heisenberg uncertainty relation, ∆Q∆ϕ 2e, which implies that the charge and the phase of the ∼ superconducting junction cannot be defined simultaneously. This leads to delocalization of the phase and to Coulomb blockade of the supercurrent, as experimentally shown by Haviland et al. [3] in the case when Josephson energy is on the order of the single-electron Coulomb energy, i.e., E /E 1. The same conclusion of delocalization applies even for J C ∼ large values of the ratio E /E [13]. J C 3 Using the differential operator due to the commutation relation, we can immediately write the quantum mechanical Hamiltonian [6] as ∂2 H = E E cosϕ. (1) − C∂(ϕ/2)2 − J When E E , charge is a good quantum number, which leads to Coulomb blockade of C J ≫ Cooper pairs and a complete delocalization of the phase. Equation 1 then takes the form of the Mathieu equation with the well-known solutions of the form Ψq(ϕ) = eiϕq/2eu (ϕ), n n where u (ϕ) is a 2π-periodic function and the wave functions are indexed according to band n number n and quasicharge q. Verification of the existence of the energy bands has been carried out by different methods [14–16]. Voltage over the junction is given by V = ∂E which changes along the energy band when ∂q quasicharge is varied. Thus, to have current flowing in the junction, the bias voltage V C (on the collector, cf. Fig. 1) has to be larger than the maximum Coulomb blockade voltage of the lowest band E : V > ∂E0 . If the current through the junction is low enough, 0 C ∂q |max dq/dt eδE /ℏ, where δE is the gap between the first and second band, the quasicharge 1 1 ≪ q is increased adiabatically and the system stays in the ground band. The junction is then in the regime of Bloch oscillations; the voltage over the junctions oscillates and Cooper pairs are tunneling at the borders of the Brillouin zone, i.e., here at q = e. Consequently, ± the current through the junction is coherent and the voltage and charge over the junction oscillate with the Bloch oscillation frequency f = I/2e, (2) B where I is the current through the Josephson junction. If the current I is not adiabatically small, we can have Zener tunneling between adja- cent energy bands. The tunneling is vertical, i.e., the quasicharge does not change. The probability of Zener tunneling between bands n 1 and n when E E is given by C J − ≫ π δE2 e I PZ = exp n = exp Z , (3) n,n−1 −8 nEC ℏI! (cid:18)− I (cid:19) where δE = E E and I is the Zener breakdown current [17–20]. Provided that n n n−1 Z − V < ∂E1 for the excited state E , the junction will become Coulomb blockaded on the C ∂q |max 1 band E after a Zener tunneling event, and no current will flow through it any more. The 1 role of the third terminal is to relax the Josephson junction back to the ground state where a new sequence of Bloch oscillations can be started. 4 B. Incoherent tunneling processes The external environment gives rise to current fluctuations that couple linearly to the phase variable. These can cause both up- and downwards transitions. The amplitude of the fluctuations is given by the size of the impedance: the larger the impedance the smaller are the current fluctuations and the transition rates. As we will see later on, the successful operation of the BOT requires one to control both the upwards and downwards transition rates. When modeling the BOT analytically, we will make use of the Zener transition rates and transitions due to charge fluctuations, both derived in Ref. [13]. The electromagnetic environment around tunnel junctions affects the tunneling process by allowing exchange of energy between the two systems [10, 21–23]. The influence of the external circuit can be taken into account perturbatively, for example, using the so called P(E)-theory [23]. A perturbative treatment of the Josephson coupling term gives rise to a result for incoherent Cooper pair tunneling [21, 23] where the tunneling electron rate is directly proportional to the probability of energy exchange with the external environment governed by the P(E) function. Taking both positive and negative energy exchange into account, tunneling both inward and outward direction leads to the total current πeE2 I(V) = J (P(2eV) P( 2eV)). (4) ℏ − − The function P(E) can be written as 1 ∞ i P(E) = dtexp J(t)+ Et , (5) 2πℏ ℏ Z−∞ (cid:20) (cid:21) which is the Fourier transform of the exponential of the phase-phase correlation function J(t) = [ϕ(t) ϕ(0)]ϕ(0) . (6) h − i The phase-phase correlation function is determined by the fluctuations caused by the envi- ronmentanditcanberelatedtotheenvironmentalimpedanceviathefluctuation-dissipation theorem. For a high-resistance environment, the P(E) function is strongly peaked at energies around E , and it may be approximated by a Gaussian function C 1 (E E )2 C P(E) = exp − , (7) √4πECkBT "− 4ECkBT # 5 where the width is governed by thermal fluctuations in the resistance R. Consequently, the subgap IV-curve displays a rather well-defined peak centered around V = 2E /e due C to the 2e charge of Cooper pairs. This characteristic feature of the IV curve provides a straightforward way to determine E of the investigated devices of small E . J J The actual downward and upward transition rates Γ (V ) and Γ (V ) as a function of in↓ C ↑ C the collector voltage were calculated by Zaikin and Golubev [24]. The Zener tunneling rate in a resistive environment, and with the assumption E E , is given by J C ≪ v v δq2/e2 Z Γ = exp 1+ h i , (8) ↑ 2τ (−v 1 " (v 1)2#) − − and the down relaxation rate due to charge fluctuations is given by v (v 1)2 Z Γ = exp − , (9) in↓ τ 2π δq2/e2 (−2 δq2/e2 ) h i h i q where v = CV /e, τ = R C , δq2 = k CT, and C C B h i π2R E 2 C J v = . (10) Z 8R E Q (cid:18) C(cid:19) The voltage v is related to the so called Zener break down current by I = ev /(4τ). Z Z Z C. BOT modeling near the onset of the bistability Our present model generalizes the previous analytic BOT theories [8, 25] by including the effect of intra-band transitions. The circuit schematics for the basic BOT modeling is depicted in Fig. 1 below. The basic circuit elements are the Josephson junction, or superconducting quantum interferometric device (SQUID) geometry, at the emitter, with a total normal state tunnel resistance of R , the single tunnel junction at the base with JJ the normal state resistance R , and the collector resistance R . The BOT base is current N C biased via a large resistor R at room temperature, but a large line capacitance C results B B in an effective voltage bias. As required by the P(E)-theory, our basic modeling is valid provided E P(2eV) 1. J ≪ The intrinsic relaxation is detrimental for BOT operation and, thus, the fluctuations should be kept low by requiring that R R = h/4e2. In practice, we need R & 100R to be C Q C Q ≫ close to the presumed idealized operation. Experimentally, this is quite hard to realize (see Sec. III) 6 Numerical analysis is needed to calculate properly the characteristics of the BOT devices near the onset of bistability. However, by introducing a phenomenological variable that describes the average number of the tunneling events N before a downward transition e h i is triggered by the base electrons [12], we may derive a rather simple description for the operation of the BOT. A value of N 1 is facilitated by intra-band transitions that e h i ≫ basically maintain the bias current of the operating point. Changes in the ratio of the bias current and the triggering current can lead to significant changes in the characteristics of the BOT. Like in theearlier analytic descriptions, the BOTemitter current can bethought ofas the result of being in either of the two states; the Bloch oscillation state with a time-averaged constant current and the blockaded state with zero current, V /R , τ = 1/Γ C C ↑ ↑ I = (11) E   0, τ↓ = 1/(Γin↓ +ΓB/ Ne ). h i The amount of time the system spends in each state is given by the Zener tunneling rate, Γ , the intrinsic relaxation Γ , and the quasiparticle tunneling rate Γ ; only every N th ↑ in↓ B e h i of the injected base electrons is able to make a downward transition. The base current, however, flows during the opposite times: 0, τ = 1/Γ ↑ ↑ I = (12) B  eΓB, τ↓ = 1/(Γin↓ +ΓB/ Ne ). h i  From these equations we can simply derive the average emitter and base currents V τ C ↑ I = (13) E h i R τ +τ C ↑ ↓ N′ I = e h ei (14) B h i τ +τ ↑ ↓ where we have defined N N′ = h ei . (15) h ei 1+ Γin N ΓB h ei By combining these two equations, we may write V τ C ↑ I = I . (16) h Ei R e N′ h Bi C h ei 7 Now, when calculating the current gain β = ∂hIEi, N′ has to be considered as a function E ∂hIBi h ei of I . Thus, we obtain B h i V τ V τ ∂ N′ β = C ↑ C ↑ h ei I , (17) E R e N′ − R e N′ 2 ∂ I h Bi C h ei C h ei h Bi which can equivalently be written as V I τ (τ +τ ) 1 C B ↑ ↑ ↓ β = h i (18) E R e2 N′ 2 1 β C h ei − H with τ +τ ∂ N′ e ∂ N′ β = ↑ ↓ h ei = h ei. (19) H N′ ∂τ I ∂τ h ei ↓ h Bi ↓ Whenβ becomes equaltoone, thegaindiverges, whichmarks thethresholdforbifurcation. H Intheregimeβ > 1, twostablesolutionsareavailableandtheoperationbecomeshysteretic H asobserved bothexperimentallyandnumerically. Hence, wemayconsiderβ asaparameter H controlling the proximity of the bifurcation threshold. For β 1, we obtain a linear dependence between β−1 and I as given by H→ E h Bi R e2 N′ 2 τ β−1 = C h ei I ǫ ↑ (20a) E "VC τ↓(τ↑ +τ↓)hIBi2(cid:16)−h Bi− τ↓(cid:17)# R τ +τ C ↑ ↓ = I +I (20b) B B−H "VC τ↓ −h i # (cid:16) (cid:17) where ǫ < 0 is a phenomenological parameter to account for the variation of ∂ N′ /∂ τ h ei h ↓i under various biasing conditions (see Appendix A). The latter term in the parenthesis of Eq. 20a specifies the threshold current I for the bifurcated, hysteretic threshold. By B−H substituting I from Eq. 14 to the prefactor of Eq. 20a, N′ 2 and I 2 term cancel each h Bi h ei h Bi other leaving the prefactor with (R /V )(τ +τ )/τ . The detailed derivation of the analytic C C ↑ ↓ ↓ formulation is outlined in Appendix B. Using a simple approximation for the variation of N′ with τ , we may derive an analytic h ei ↓ formula for the bifurcation threshold on the E vs I plane (see Appendix A). The E J B J h i dependence of I comes mainly from Eq. 8 which leads to the analytic form given by, B−H I Γ +exp( κE2) B−H se↑ − J (21) e ∝ 1+Γ2 /E4 B J q where the first term in the numerator, Γ is the upward transition rate due to single se↑ electron tunneling, whereas the second term arises due to LZ tunneling. The parameter κ 8 involves all the other parameters inside the exponent of Eq. 8. This functional dependence between I and E in Eq. 21 is also in good agreement with the results of our numerical B−H J simulations. The BOT behaviour described here is referred to as ‘normal’ operation. In this configu- ration, the junction is initially in the upper band and a quasiparticle tunneling due to base current will bring the junction to the lowest band where it performs Bloch oscillations. This coherent oscillation will be inhibited by Zener tunneling and the system jumps back to the upper state and the whole process is repeated again. If the sign of V (and consequently C I ) is reversed the base current will induce transitions to the upper band, an operational E mode that we call ‘inverted’ operation. Since the ‘normal’ operation is conceptually clearer we have concentrated our studies in this mode of BOT. II. FABRICATION AND MEASUREMENT The BOT samples employed in this work were fabricated using a 20-nm-thick Ge mask on top of LOR 3B resist. Patterning of the Ge layer was performed using conventional e-beam lithography at 20 keV. After patterning, the PMMA layer was developed in MIBK:IPA (1:3) solution and subjected to a plasma etch with CHF plasma. Finally, the LOR under the 4 germanium was etched in oxygen plasma up to the desired extent of undercut. Shadowangleevaporationatfourdifferent angleswasemployed togeneratethestructures consisting of three metals. Originally, the BOT was envisioned to have a NIN junction as the base junction, but the technique of manufacturing both SIS and NIN junctions on the same sample is exceedingly difficult and, therefore, we opted to have a NIS base junction instead. The SIS junction is formed of two Josephson junctions in the SQUID geometry; this facilitates tuning of the Josephson energy by magnetic flux. The process order in the evaporationsequence was(I)Chromium, (II)Aluminium, (III)oxidization, (IV)Aluminium, and (V) Copper. NMP or PG remover was used for lift-off. Oxidation was done in Ar:O 2 (6:1) mixture at 80 mTorr for 1 min. A typical sample used in the present study is displayed in Fig. 1. The area of the SIS junctions is 100 x 150 nm2 each (equal areas within 10%). The NIS junction on the base has an area 70 100 nm2, roughly half of the SQUID junctions. × The measurements were done on a plastic dilution refrigerator (PDR-50) from Nanoway 9 C I C R C R N B Q I I B R JJ E (ϕ) J I E E FIG.1. Scanningelectron micrographof thesample(left frame)andaschematic viewof thedevice (right frame). Inboth pictures, base, emitter and collector are marked by B, E and C,respectively. Positive directions for the currents are indicated by the arrows. The sample parameters are given in Table I. Q (t) is the island charge tracked in the numerical simulations. I BOT # R R R E Emin E ∆ N JJ C J J C 1 53 27 550 17 2.7 40 150 2 75 21 305 25 3.3 60 165 TABLE I.BOTparameters for themeasuredsample. R andR arethenormalstate resistances N JJ oftheNISandJJtunneljunctionsintheSQUID-loopgeometry, respectively. Resistances aregiven in units of kΩ and energies in µeV. Ltd. Thebase temperatureof therefrigeratorwas 50mK.Thefiltering inthePDR consisted of 70 cm long Thermocoax cables on the sample holder and 1 kOhm serial resistors at 1.5 K. In addition, micro-wave filters from Mini-circuits (BLP 1.9) were used at top of the cryostat. The measurement set-up in this work was similar to that described in Ref. 26. The BOT base was DC current biased by a resistor R =1 10 GΩ, which was located at room B − temperature. Voltages were measured with low noise LI-75A voltage preamplifiers while currents were monitored using DL1211 low noise current amplifiers. The resistance values of the three circuit branches were determined at 4.2 K. Since there was a weak temperature dependence in R , we determined the actual value from C 1/√V asymptote [27] of the IV curves measured at low E . The maximum Josephson J energy E was calculated using the Ambegaokar-Baratoff relation which yielded E = J J 10