PHYSICAL REVIEW B

VOLUME 26, NUMBER 1

1 JULY 1982

Measurements of quantum noise in resistively shunted Josephson junctions
Roger H. Koch, D. J. Van Harlingen,* and John Clarke
Department of Physics, University of California, Berkeley, California 94720 and Materials and Molecular Research Division, Lawrence Berkeley Laboratory, Berkeley, California 94720 (Received 19 November 1981)

Measurements have been made of the low-frequency spectral density of the voltage noise in current-biased resistively shunted Josephson tunnel junctions under conditions in which the noise mixed-down from frequencies near the Josephson frequency (vJ) to the measurement frequency «<vJ) is in the regime hVJ>kBT. In this limit, quantum corrections to the mixed-down noise are important. The spectral densities measured on junctions with current-voltage characteristics close to the Stewart-McCumber model were in excellent agreement with the predicted values, with no fitted parameters. The mixeddown noise for a wide range of bias voltages was used to infer the spectral density of the current noise in the shunt resistor at frequency v. With no fitted parameters, this spectral density at frequencies up to 500 GHz was in excellent agreement with the prediction (2hv/R)coth(hv/2kBT). The presence of the zero-point term, 2hv/R, at frequencies h v > kB T was clearly demonstrated. The current-voltage characteristics of a junction with f3L =21TLs10/<I>o-1 and f3c =21T1 oR 2C /<1>0« 1, where 10 is the critical current, C is the junction capacitance, and L, is the shunt inductance, showed structure at voltages where the Josephson frequency was near a subharmonic of the LsC resonant frequency. The additional nonlinearity of the 1 V characteristic caused mixing down of noise near higher harmonics of the Josephson frequency, thereby greatly enhancing the voltage noise. The measured noise was in good agreement with that predicted by computer simulations. I. INTRODUCTION The effects of thermal noise on a resistively and for frequencies much less than v;, the spectral density of the voltage noise is given by S (0)= 4kBTRE
v

shunted'v Josephsorr' junction (RSJ) have been extensively studied. The theories assume that the noise originates as Nyquist noise in the shunt resistor R. The junction is modeled as a particle moving in a tilted periodic potential, and the effect of the noise current is to induce random fluctuations in the angle of tilt. These fluctuations have two effects. First, they enable the phase of the junction to slip by 21T when the bias current 1 is less than the noise-free critical current 10, thereby producing a voltage pulse across the junction. This effect produces noise rounding of the 1- V characteristics at low voltages, V; the noise rounding has been calculated by Ambegoakar and Halperin" and Vystavkin et al? for the case c=o (C is the capacitance of the junction). Subsequently, Kurkijarvi and Ambegoakar" and Yoss" computed the case C::;i=0. Second, the fluctuations generate a voltage noise when the junction is current biased at a nonzero voltage. Likharev and Semenov'' and Vystavkin et al.s showed that for the c=o case in the limit hVJ «kBT (vJ=2eVlh is the Josephson frequency) 26

R

I

1+-I 2 1

[10 ]2]

.

(1.1)

Here, R D is the dynamic resistance. This result was derived on the assumption that the noise is sufficiently small that one can neglect departures of the 1- V characteristic from that of the ideal RSJ,I,2
V=R(I2-I5)1I2. (1.2)

Thus, Eq. (1.1) is not valid in the noise-rounded region 1</0' Voss" and Koch and Clarke" computed the noise for the case C¥=O. Experimental results are in good agreement with calculations for both the noise rounding!" and voltage noise. II For a junction voltage-biased on a self-resonant step, Stephen 12 has calculated the contribution of pair current fluctuations to the linewidth of the Josephson radiation. This noise arises from photon number fluctuations (including zero-point fluctuations) in the lossy cavity formed by the junction, and is not intrinsic to the tunneling of Cooper pairs in a nonresonant junction. Experimental re74

Section IV contains some concluding remarks. The contribution of noise-generated near frequencies 2vJl3vJl . thus generating a randomness in the rate at which the phase B propagates. as has been calculated by several authors.4) In the limit1.. In Sec. In this paper we describe measurements of the voltage noise in current-biased overdamped junctions (Pc < 1) in the free-running mode I> 10. and the noise current In{t) has a spectral density'? Sj(v)= 2~V coth [ 2~:T [ ] 4hv =T exp(hvlkBT)-l 1 + 2" 1] (1. (1. More recently.1. This model also predicts "quantum activation. Equation (1. i. which has some probability of penetrating the barrier by macroscopic quantum tunneling (MQT). Koch et al. is negligible in the ideal RSJ model. (1.hI2e).04 pF . (1. where (1. III we present the experimental results and compare them with the predictions of the theory. reduces to Sv(O) ture. the spectral density of the voltage noise Sv(O) is given by Sv(~) RD = 4kBT R + 2eV R [10 ]2coth [~ ].the first term on the left-hand side of Eq. (1.. An appendix outlines the procedure used in the computer modeling. The equation of motion for the junction is 1iC + 21iR B+losinB=I 2e e s +IN . = lO4 Acm-2. 250-nm-thick Ph (20 wt % In) film.um-wide Cu (0-3 wt% AI) film 40. II we describe the experimental procedures and in Sec. we find K:::::. In our pic- .2 K. quantum corrections to the mixed-down noise become important. Instead.21TloR 2C/<I>o« 1 (<I>o=. At 4.3) in which the particle is a point mass is likely to become invalid. In the ground state at T=O.um-2.5) reduces to Eq.f At frequencies much less than VJ and in the limit 1110> 1. we suspect this description will fail when I becomes significantly smaller than IQ. Our junctions [Fig.. and the equations can then be solved analytically using the techniques of Likharev and Semenov.i" We first deposited a lO-. EXPERIMENTAL PROCEDURES A. 1(a)] were fabricated on glass substrates using photolithographic lift-off techniques. Writing K=(e IkBT)(Pc<l>oh 121Td/2.generated in the shunt resistor become important. where i 1 is the critical current density and c is the capacitance per unit area. I kBT (1. transitions out of the zero-voltage state become very infrequent.um-wide.7) and the observed noise is generated solely by zeropoint fluctuations in the shunt resistor.6) In the extreme quantum limit eV»kBT. and c=O." that is. while the second term is noise-generated near the Josephson frequency that is mixed down to the measurement frequency by the nonlinearity of the junction. Eq.2 0 < Pc =. the zero-point energy still induces fluctuations in the tilted periodic potential.4He temperature range. II. Junction fabrication To observe quantum noise effects.3) can be neglected.3) where B is the phase difference across the junction.16-19 Clearly. a quantitative theory that can deal with both zeropoint fluctuations and MQT for all values of the bias current and damping is very much needed. with Pc =0.5) The first term on the right-hand side is noisegenerated at the measurement frequency.5) when eV(IolI)2> 2kBT. 1. (1. 75 sults'! are in good agreement with the predictions. the resistor can be modeled as a large collection of harmonic oscillators.5) R~ 2eV =T [10 ]2 T (1. in which noise rounding can be neglected... where there are no thermal fluctuations. and the model represented by Eq.to lOO-nm thick and then a to-.V characteristic of an overdamped junction (Pc < 1) even at T=O due to zero-point fluctuations. Although we cannot yet make any quantitative statements.14 considered the limit h VJ > kB T in which quantum correction to the nois. and the second term will become com parable to the first term on the right-hand side of Eq. one must treat the particle as a quantum-mechanical wave packet. we see that junctions with high critical densities are necessary to observe these effects in the liquid.2.26 MEASUREMENTS OF QUANTUM NOISE IN RESISTIVELY . These requirements can be met provided K?_ 1. In this limit. (1. we require junctions with K?_ 1.1) in the limit eV «kBT. In the limit eV?_kBT. a noise-rounding of the 1.

V characteristic and dynamic resistance and obtained the shunt resistance by reducing the critical current nearly to zero. After further amplification. rPb1n Bose Electrode AND CLARKE 26 I Pb Counteretectrode-" Cu. The two cooled LC-resonant circuits with inductors LIl. (i) Corrections were made for the measured preamplifier voltage and current noise. which contained leads that were partially at room temperature.1 to 2 mA at 4. the error in the correction was ±5%. Each tank circuit was connected in turn to a Brookdeal 5004 preamplifier. to about 0. The preamplifier noise was comparable with the junction noise at 4.7 O. and a final protective layer of SiO was evaporated. (b) schematic of measuring circuit. The critical current ranged from 0. and surrounded by a superconducting can and a mumetal can.2 K./' 10!'-m (0) f--I (b) FIG.5.5-k 0 resistor RF and the cable capacitance Cc. In the last case.1 kO) to ±2%. The system gain was calibrated against the Nyquist noise of a resistor Rc(5. in addition. The circuit for measuring the noise across a junction is shown in Fig.76 Pbln -]n203 -Pb Tunnel Junction I : KOCH.05 to 0. about 106 kHz. (a) Configuration of resistively shunted tunnel junction. intermediate frequency.i! For example. III D). (ii) Noise due to losses in the tank circuit was negligible for the 70. Lt2 and capacitors Ctl> Ct2 had resonant frequencies of 70 and 183 kHz. The quasiparticle current Iqp contributes a noise with a current spectral density+' 2e1qpcoth(eV/2kBT). for Pc appreciably greater than zero (junctions 2 and 3). which we estimate to be < 10-2 at 4. lfb). (v) The power dissipation in the shunt resistor . reducing its inductance L. (iii) From measurements at three frequencies we determined that some junctions (2 and 4) generated a small amount of 1// noise. Measurement procedures Each junction was immersed in liquid 4He.9 Thus. (iv) The noise measurements were performed at bias voltages well below the sum of the gaps of the two superconductors.AI Shunt__. the noise was mixed-down to frequencies below 500 Hz and its spectral density was measured with a typical averaging time of 10 min. Thus. After patterning the resist for the upper electrode. while the value of Sv (0) does increase significantly as Pc is increased in this range. by connecting together the tank circuit leads we could measure the noise at a third. the ratio of the spectral densities of the quasiparticle and mixed-down noises is of order 1qp/(V/R). the error introduced was no more than ±3%. was deposited and two windows were opened by lift-off to expose the Pbln and CuAI films. We measured the 1.and 183-kHz tank circuits.2 pH. The low-pass filters for the bias current consisted of a cooled 1. We now discuss the various measured corrections to the noise spectral density. The Pb counter electrode formed a ground plane for the shunt. it is more appropriate to compare experimental and theoretical values of Sv(O)/R~ rather than Sv(O).5 pF (see Sec.5 usx: and its capacitance was about 0. so that the required quantity Sv(O)/R~ was independent of Q. 100-nm thick. The diameter of the junction was about 2. the dashed lines enclose the components immersed in liquid helium. even if the uncertainty in the noise was as high as ±30%. Junctions fabricated with these techniques omitting the resistive shunts displayed excellent tunneling characteristics with little excess current at voltages below the sum of the gaps.:s . B. We note that the predicted value of Sv(O)/R~ is virtually independent of Pc in the range 0 < Pc 0.2 K.7. but not for the 106-kHz tank circuit. the In203 was grown thermally in a low pressure of oxygen. the 400-nm-thick Pb counter electrode was deposited and lifted-off. l. A SiO layer. and the corresponding error introduced by the correction was about ±5%. for junction 2 at 183 kHz the 1// noise was typically 10% or less of the white-noise spectral density at the higher voltages. Van HARLINGEN. the exposed metal surfaces were cleaned by rf sputter-etching in Ar. The noise produced by the junction across the tank circuit was Q2Sv(0)=(V2L?[Sv(O)/R~].2 K and the resistance of the shunt ranged from 0.

so that the 1.77 n at 400 ttV 0.67 n at 50 tt V 0. and measurements on junctions in the classical limit eV «kBT showed the correct temperature dependence and were in excellent agreement with theory (see Sees.99 6. Parameters of junctions. 0. III A.26 MEASUREMENTS OF QUANTUM NOISE IN RESISTIVELY . 77 caused a significant temperature rise at the high bias voltages in some junctions. (3.21 0.2 1.5x 10-22 <3x 10-24 183 kHz 100 kHz 7 1.22 1. The 1// and heating corrections were negligible through the range of measurement. A.0 3.2 pH for 1.20 0. Junction Temperature 10 (mA) R (11) (K) 4.36 1.n at 100 tt V 0.n at 200 tt V 0. We determined the rise aT by reducing the critical current almost to zero and measuring the Nyquist noise of the shunt as a function of power dissipation.62 1.23 pH for 4. IV D for a discussion of the effects of the value of f3L). Junction 1 As a test of our measurement system and of the effectiveness of the shielding we first investigated a junction in the limit K «1 in which the LikhartwSemenov'' result.2.032 0.4 0. For most junctions the heating effect was important only at bias voltages V»kBT/e. where the heating correction was particularly large. so that the only corrections to the measured data were for preamplifier and tank circuit noise. the measurements were at two frequencies only." and the predicted spectral density of the noise diverges as 1--10. Measurement of the Nyquist noise in cooled resistors were within ±3% of the predicted value. 2 we compare the measured noise with the predictions of Eq. EXPERIMENTAL RESULTS AND COMPARISON WITH THEORY We report results on four different junctions that illustrate various aspects of the theory.38 0. At 4.003 and 0.V characteristic was very close to that for an ideal resistively shunted junction (see Sec. I.17 5. and T were used.066 <2x 10-22 100 kHz <1 4.003 0. is applicable. tial parameters of the junctions are listed in Table I.084 n at 50 tt V 0. it was sufficient to correct the data by subtracting 4kBaT/R from the measured noise.1).51 0. for junction 3. the value of K was 0.5 pF for 1.70 . In plotting the theoretical points we used the predicted dynamic resistance'r' RD=R [ I-loll 22]-1/2 .05 0.58 n at 50 tt V 0. 10.45 0. The essen- so that only the measured values of R..2. Eq. 30 and 100 kHz.75 . 0. (1.n at 400 ttV 0. we believe our measurements were not significantly influenced by extraneous noise sources.1).6 .37 3.3.2 0.60 3 4 e. (vi) We took considerable care to shield the experiment from extraneous noise sources. L =0.3..62 .2 K.25 0.5) is nearly independent of the shunt temperature.2 0.68 . (1.6 0.53 0. (In this experiment.1) III. However.0X 10-23 183 kHz 1.066. Thus. where the mixed-down term in Eq. The parameters f3c and f3L =. the agreement between theory and experiment is very good indeed.32 0.21TLslo/<I>o were 0. and III C).0X 10-23 183 kHz 0. III B. At low bias voltages the measured noise agreed with the Nyquist formula to within ±3%. f3L (A2Hz-1) Frequency Heating (K/ttW) si" K 0.6 2 1. Thus.) In Fig. (1. Above the noise-rounded region. C =0. The good agreement between theory and TABLE I. the maximum error introduced was ±3%.n at 100 ttV 0.31 0. At very low voltages.092 n at 100 ttV 0. we have neglected noise rounding. the measured noise decreases as the current is lowered because the noise-rounded dynamic resistance decreases. Thus. the value of R at 100 ttV was used to compute f3c and K. respectively. it was necessary to correct the mixed-down term was well. and to avoid coupling significant 3OO-K noise into the low-temperature circuitry.81 pF for 4.2.075 4.

38 and 0. (1. solid line is prediction of Eq.(O) vs I for junction 1 at 4. In the mid-voltage range.2 K.5) using the measured values of R. about 0.0 at 1.6 K (the critical current increased slightly as the temperature was lowered).75 n as the voltage bias was increased from 0 to 1 mV.5) and (3. .8 __ L-~·.].2 K to 3. Another deviation from the simple model arose because the shunt resistance R which was measured with the critical current suppressed nearly to zero. varied between 0.99 at 4. The solid line through the solid circles is the prediction of Eq. (1. both corrections are small.5) is not affected by the additional nonlinearity. and T. 3. Solid circles are data with dashed line drawn through them. The upper dashed line is the predicted noise in the absence of zero- 4. The solid circles are the noise after the 1/f noise subtraction and the heating correction have been made. (1. 4.78 KOCH. 1. V. 0.65 and 0.2). while at high voltages the correction is largely due to heating.0 FIG. (1. Van HARLINGEN. S. 4 we plot measured values of Sv(O)IRJj versus voltage (open circles) after the preamplifier noise has been subtracted. At low voltages the correction is entirely due to 1/f noise. The values of [3e and [3L.5) and (3. . Junction 2 The parameters of this junction (Table I) were chosen to emphasize the quantum effects: Thus K increased from 0.2 K. Solid triangles are measured mixed-down noise. 10 6 2 10·L_L_~~~~_L_L~~~2~L-L-~ I (rnA) FIG. We believe this variation was the result of a proximity effect between the shunt and the electrodes or possibly of diffusion of Pb into the shunt.6 V (rnv ) 0. ]0. at 183 kHz vs V for junction 2 at 4. were small enough that the deviations from the ideal resistively shunted junction were relatively minor. In Fig.4 0. solid circles below show the noise remaining after correction for 11 noise and heating. Figure 3 shows] and dVld] vs Vat 4. respectively. S.31 at 4.1).1) will be suppressed provided the mixing coefficient (]fl212) in Eq.2 K. We emphasize that in comparing the quantity Sv(O)IRJj with the theory. The measured value of R was used at each voltage bias when we compared theory and experiment.2 0. 2.2). (3. There is a small drop in dVld] at about 800 /LVwhich we believe is associated with a resonance of the shunt inductance and the junction capacitance (see Sec.(O)IRi. III D).2 K. lower solid and dashed lines are mixed-down noise predicted by Eqs. I and RD vs V for junction 2 at 4. AND CLARKE 26 600 200 I 100 ~60 " N g s V(mVI (/) -. experiment for] >]0 indicates very strongly that the contribution of extraneous noise sources is negligible.2 K OL_~--~--~~~~~~~~~ o 0.2 K. B. There is also some very fine structure and a dip at 300 /LVof unknown origin. The open circles show the total measured noise across the junction.--Measurement Theory FIG. f Upper solid and dashed lines are predictions of Eq. small deviations in RD from Eq. 20 --.

(1. At 50 /L V. The mixed-down noise at 350 /LVand above is independent of temperature. The solid lines are the corresponding predictions of Eq.2).. (1. and remains in good agreement with the predictions of Eq.: . In Fig.7). (1. (1. The mixed-down noise at the six highest voltages is in the quantum limit at all temperatures measured. (2eV /R)(10/J)2coth(eV /kBT) . lie somewhat above the prediction. 5 we show the temperature dependence of the noise for twelve bias voltages ranging from 50 to 550 /LV.~ 4 t 0' c 0/ 0/. the mixed-down noise is in the classical limit for the whole temperature range and proportional to T as expected. 10 increased slightly. The data at 300 /LV. (4eV /R)(10/J)2[exp(2eV /kBT)-l]-1 . ~c ' L 200". (1. 00 2 4 6 0 2 o 4 6 T (K) FIG..V .' O~"'/ ~ O/~/.' ./ . 5. and are substantially higher than the predictions of a theory that does not include this contribution.) As the voltage is lowered the mixed-down noise becomes increasingly temperature dependent. The solid line through the triangles is the mixed-down noise predicted by Eq. 6 for 4. (3. Arrows indicate 2e V = kB T. 4.5).26 MEASUREMENTS OF QUANTUM NOISE IN RESISTIVELY .. the total junction noise is in good agreement with the predictions of Eq.2 K (solid circles) and 1..5). giving rise to the slight increase in the mixed-down noise that is evident in both the data and the theoretical prediction. 4 represent the measured mixed-down noise.5) at very low voltages are possibly due to our neglect of noise rounding in the theory. '.--'--' ~ The small discrepancies between the data and Eq. (3. . HID). (1... that is S~(o) 50". mixed-down noise at temperatures well above this temperature is in the classical limit eV «kBT.. (1.4) using measured . The results are plotted in Fig./ . ~- -.V R~ 4kBT =~ +R T 4eV [10 12 -e-xp--:(-2e-V-/':""k. 79 point fluctuations. while that at temperatures well below this temperature is in the quantum limit eV»kBT. Sv(0)/R~=(2eV /R)(Io/I)2 ..• :It • 7/ oo~ . ** 4 6 ' P 550pV . 8~~~-450pV 6 500pV 4 2)// oo~ /~K . O~~"'--I...2) in the absence of zeropoint fluctuations.6 K (open circles). -...2) The triangles in Fig. . Notation is as for Fig.5) into Eq.4). 2). The temperature T=2eV/kB is indicated for the six lowest voltages.5) and substantially greater than the predictions of Eq. Sv(O)/R~ at 183 kHz vs T for junction 2 at 12 bias voltages. The notation is the same as that in Fig. and in excellent agreement with the value of Eq.however.. (1. 3) that increases the magnitude of the mixed-down noise above the value predicted by Eq. ~. o iO ' / . (As the temperature was lowered. / . which was computed by subtracting 4kBT/R from the solid circles.. We can extract from our data the measured spectral density of the current noise Sj(v) generated by the shunt resistance R at the Josephson frequency v=2eV/h. (3.6 <f) > 4 2 o· . a procedure that converts the mixeddown noise in Eq. 4. . .5) (this topic will be discussed in detail in Sec..-T-) --1 B 6. This discrepancy arises from the structure at 300 /LV (see Fig. We divide each value of the mixed-down noise by the mixing coefficient (10/1)2/2. For all twelve voltages. (1. It is evident from Fig. + . This temperature dependence demonstrates that the contribution of any extraneous noise was negligible.' a l l . . 4 that both the total measured noise across the junction and the measured mixed-down noise are in excellent agreement with the theory that includes a contribution from the mixed-down zero-point fluctuations..------ ~ 300JLV while the lower dashed line is the mixed-down noise predicted by Eq.

6 K (open circles).8 10=0.07 ~ 10. the three lowest values were obtained by trapping flux in the junction.5l.65 to 0. values of v=2eVlh. the equivalent circuit is shown in the inset in Fig. while dashed lines are 0.80 KOCH. Junction 3 Junction 4 An alternative means of varying the mixed-down noise between the quantum and thermal limits is to change 10 at fixed temperature. the correct observed dependence of the noise on 10 demonstrates the absence of any significant extraneous noise. and we have investigated its origin and its effect on the noise in some detail. 7 we plot Sv(OlIR~ vs Vat 4.34 N"" N N <. 9. the dashed lines represent the theoretical prediction in the absence of the zero-point term.2 K (solid circles) and 1. FIG. the measured total noise and the measured mixed-down noise are in very good agreement with the predictions.V and (dVldD.2 K o (a) i. By contrast.--.65. Once again.4). (1. the data lie convincingly above the theory that does not include the mixed-down zero-point fluctuations.--. Notation is as for Fig. :r: N o O~_L __ ~ __ ~~ (e) __ ~ (d) Io = 0. For K=0. while for K=0. R. The structure arises from the resonant circuit formed by the shunt inductance L. (1. Sv(O) at 183 kHz vs V for junction 3 at 4. The existence of zero-point fluctuations in the measured spectral density of the current noise is rather convincingly demonstrated. some junctions contain resonances that can effect the magnitude of the noise mixed-down to the measurement frequency. 6.6 K 8 /{ 0. The resonant circuit pulls the Josephson frequency slightly so that it become more closely a subharmonic of the resonant frequency.--~ NO rr :::. Junction 4 exhibited strong resonant structure.--. Measured spectral density of current noise in shunt resistor of junction 2 at 4. bearing in mind that.V characteristics at 1. 26 • 4.--. C.095 mA /{ = 2 ()) 017 io" 11 (Hz) > FIG. Apart from this discrepancy. The 1// noise in junction 3 at 183 kHz was insignificant k 2 % l. and junction capacitance C. Solid lines are prediction of Eq.---.65 K = 0.--.038mA K= 0. no fitting parameters are used.V characteristic.07. the dynamic resis- .-.--. In Fig.2 K for four values of 10. so that it was necessary to correct the mixed-down noise in addition to the noise generated at the measurement frequency. and fall far below the data at the higher frequencies.--. The critical current was lowered by trapping flux in the junction. 4. AND CLARKE 10. the presence of a resonance near 200 11 V increased the magnitude of the measured noise As noted earlier. (4hvIRl[exp(hvlkBTl-1]-1.--. and T.2 K for four values of 10 corresponding to values of K ranging from 0. The agreement between the data and the predictions is rather good. but the heating correction at the higher voltages was substantial. At the highest two values of 10. as the current bias is increased. =0365 mA = (b)Io = 0. somewhat above the prediction of Eq.191 A m 1.1 K for four values of critical current. 7. once again. Hence. D. Figure 8 shows the 1.07 the contribution of the zero-point term is less than our experimental error. Van HARLINGEN. The slight increase of the data above the theory at the highest voltages may reflect the presence of a resonance on the 1.2 V(mV) (4hv/R)[exp(hv/kBTl-lj-l.

we conclude that the 1.6 kHz. To obtain these curves. by R with spectral density 2hvR coth(hvI2kBn.1 8. 98.6 V (mV) 1. The agreement between the measured and computed values is quite good.V characteristic will show substantial resonant structure when (3L = 21TLsIo/<l>o ?: 0. using L. it was necessary to fit the values of L.68 mA fJL ~047 fJC • 0.2 0.4 K with 10= 1. a fact that could not readily be included in the computer simulation (see Appendix). I and RD vs V for junction 4 at 1.V characteristics and the spectral density of the voltage noise across the junction. I-V and dV /dl-V curves for junction 4 at K for four values of 10. the measurements lie slightly below the computed values at lower voltages.. 0. K= 1.3) and (3. and we measured the noise mostly at one frequency only. RO 01 (e) 10 • 0. R =0. This junction was investigated at an early stage of our work. using the procedure outlined in the Appendix. although the measured structure at the higher voltages is considerably more smeared than predicted. even though noise rounding is negligible in this region. 8.4 K. This discrepancy occurs because the measured shunt resistance at low voltages dropped somewhat below the high-voltage value. and C.17. and /3c =0. =0.23 pH and C=0. possibly because of noise rounding. From our simulations. with a few measure- .81 pF. 81 5 4 02 -L-l_~10~0~~~~2~0~0~~~~3~00~-~0 V (jJ-V) (b) !O'120mA (mA) ~~~==~~~==~~~==~~O fJC' 0.025 (_fl.. The computed dynamic resistance is also shown.05. the temperature at which the noise measurements were made.4) where Is is the current flowing through the shunt and VN is the equilibrium noise voltage generated Figure 9 shows I and RD vs V for junction 4 at 1. 1.014 0. 9. these values are consistent with values expected from the dimensions of the sample. 8 and 9 in is an integer). /3L = 1.032.4 0.092 fl.26 MEASUREMENTS OF QUANTUM NOISE IN RESISTIVELY . We have computed the 1. Furthermore. The equations of motion are (3.1 tance will be alternately increased and decreased as the Josephson frequency passes through each subharmonic frequency of the LCR resonance. Inset is equivalent circuit of junction._) fJL' 0.0 o FIG.82 FIG.53 rnA. The lin dependence of the dynamic resistance is shown clearly in Figs. Dashed line is computed RD. 5 and the approximate Q of the LCR circuit ((3L l(3c) 1/2 »1. The observed rapid decrease in the magnitude of the resonant structure as lois lowered is demonstrated in Fig.

6) from the data at 98.1) for voltages well below kBTle can be written as (2kB TR)(JoRIVJ2. l lta).--'-__ __ 10 -\ kHz __jL_. We define a mixing impedances Zk via the re- . We then assume that the spectral density of the excess noise scales as 1/1 and from data at the two voltages where measurements were made at two frequencies we calculate the spectral density of the 1II noise in the critical current: SJ:f(98. typically 15% or less of the total junction noise at 200 JLV.-'::.&. The agreement between the measured and computed values is now rather good.6 kHz. At voltages where the RSJ result. Triangles are measured mixed-down noise assuming excess low-frequency noise is proportional to 1/ f.6 kHz at several voltages between 100 and 200 JL V and obtained values that were consistent with those obtained by the above procedure.P the spectral density of the voltage noise should be proportional to (avlaJo)2._j. We used the following procedure to subtract the 1II noise in the range of voltage where the oscillations occurred. with the zero-point term included and with the values of L. Our computer simulation yields the magnitudes of the contributions of the noise generated at multiples of the Josephson frequency as shown in Fig. is valid we find (3.2). the mixed-down noise in Eq.98. By subtracting the 1II voltage noise computed using Eq. and C obtained by fitting the model to the J. Figure 11(a) shows the spectral density of the measured voltage noise at 1. First. '4.82 KOCH.----.5X 10-22 A2Hz-l. which we estimate to be ±6% of the total spectral density.6 L-. (1. The mixeddown noise is in excellent agreement with the predicted value.6 kHz the plotted quantity 10 r--r--. As a result the heating correction had a higher uncertainty. U1 •. In Fig.____j 100 FIG. Solid lines have slope .6 kHz and for two voltages at 31.2.6 kHz. we obtain the mixed-down noise shown in Fig. if the 111 noise arises from fluctuations in the critical current.--'. Spectral density of total voltage noise across junction 4 at two frequencies in the region V < k B T / e with (4kBTRb/R) subtracted out (open and solid circles). this procedure provides strong evidence that the spectral density of the excess noise at low voltages scales closely as 1II (as is the case for all junctions on which we have measurements at three frequencies).__ . The solid line shows the result of the computer simulation. As a further complication. Van HARLINGEN. Second.l. The data tend to lie somewhat above the computed curve at voltages above 100 JLV. suggesting that the 1/1 noise scales as WVlaIo)2.. 4k8TR 2 10 -23 (IOR)2~~ V . (3.6 kHz using the value of SJ:f quoted above.V characteristics in Fig. we did not measure the heating correction on this junction.4 K. scales with 1/V2. together with measured values of aVlaJo. AND CLARKE 26 ments at 31. 11(b)we have applied a heating correction by subtracting 4kBf1TR~/R from the solid circles in Fig. Thus. 10. Eq. Figure 10 shows the spectral densities of the voltage noise across the junction for V < 100 JLVat 98. with the direct term (4kBTIR)R~ subtracted out.' . -21 " ~ I > -22 10 > . R. 12.5) Hence.1).6 kHz. than for the other junctions. but rather on one fabricated simultaneously. Thus. dashed line is prediction of Eq. (3. 10.6) where SJ:f(v) is the spectral density of the 111 fluctuations in the critical current at the measurement frequency. indicating that our model is a good approximation. with the 111 noise subtracted.-. We then calculated the 1II voltage noise at higher voltages (> 100 JL V) from measurements at 98. (1.---.-:-___. at low voltages where the deviations from the RSJ model are negligible and for fixed values of J 0.---. Since the overall 1/1 correction was small. 10. the voltage noise arising from 1II fluctuations will be S~/f(v)=R2(IoRIV)2SJ:f(v) . At 98.---. We also measured the noise at 31. the spectral densities of both the mixed-down noise and 1/1 noise ( and their sum) should be proportional to 1/V2. we believe that the error introduced by the correction is at most ±5% of the mixed-down noise. and T. together with the measured mixed-down noise computed by subtracting 4kBTR~/R. (1.6 kHz)=5.

4. These results explain quantitatively the additional noise associated with the resonant structure.. (1. 1. 0 2 r: 10 -4 :3 4 5 k 6 7 8 m 9 10 FIG.031 for three values of f3L and a bias current 1110=1.4 and 1. and the noise is considerably enhanced over the value predicted by Eq. For =0.05. 2. the additional noise observed on junctions 2 and 3 in the vicinity of structure on the 1. We emphasize that in comparing the data for junctions 1. and qualitatively.. (k) (0)= ~ 00 k=O I z. evidence that these fluctuations give rise to the limiting voltage noise in a current-biased resistively shunted Josephson junction in the quantum limit for I> 10 (Figs. e. 2.031 and PL =0. IZkI2IR2forajunctionwithf3c=O. z. In fact. and the deviations from the RSJ model are negligible. 1 2S/(kvJ) .05 10 2 r: V. Although the data obtained from junction 4 are considerably harder to interpret than those from the other junctions. On the other hand. junctions 2 and 3 provide the main evidence for the accuracy of Eq. . We note that I 12=R~. the role of zero-point fluctuations is even more important because of the large number of harmonics that contribute to the mixed-down noise. 10 -2 '" (c) !3L' 1. (0) !3L' 0.42. 6).7) o ~ :i! 8 '" 'i' > o o 4 o o o 0 o o o o o ~ J:~ OL__L __ ~~ __ _L~~~==~~=-~~~~ o 40 80 120 V (pV) 160 200 FIG. Solid and f dashed lines are predictions of computer simulation with and without zero-point term.05 - where k=O.4 K. and S~k)(O) is the spectral density of the mixed-down voltage noise due to noise near frequency kVJ. The noise generated at frequencies near the higher harmonics can be in the quantum limit even for junctions with K < I. IV. the good agreement between our results and Eq.5). solid circles are mixed-down noise with 11 noise subtracted. (al Open circles are measured voltage noise at 100 kHz across junction 4 at 1.5). there is no fitting of the data.. 83 lation (see the Appendix) (0 12 o l Sv(O)= ~ 00 k=O s. (b) Solid circles are data after heating correction has been made. CONCLUDING REMARKS n: '"<.5) justifies our use'" of a Langevin equation together with a zero-point driving term to predict quantum noise effects in a current-biased Josephson junction in the overdamped limit when it is in the free-running mode 1 >10' We were not able to examine the validity .. Furthermore. first of the existence of a zero-point term is the spectral density of the current noise of a resistor in thermal equilibrium (Fig. and 7). and 3 with theory we have used only measured parameters. z. Thus. is essentially zero for k 2 2. 12. respectively. solid line is identical to that in (a).05. We believe the results obtained from these junctions are a convincing demonstration. there are very substantial contributions to the noise from harmonics out to the fifth and ninth. and second. (3. 11. (1. 5. (1.V characteristic.26 MEASUREMENTS OF QUANTUM NOISE IN RESISTIVELY . for PL =0. the capacitance and inductance of these two junctions were estimated from computer fits to this structure.

this method appears to account for most of the data observed on junction 4 satisfactorily. Van Harlingen would like to thank the National Science Foundation for preand post-doctoral fellowships. which exhibited resonant structure. B. Office of Energy Research. however. J. Jackel and his colleagues for helpful discussions on junction fabrication. a value at least five times greater than the observed mixed-down noise at 1 mV. f3c ::= 217IOR 2C/<I>oand f3L ::= 217LJ 0/<1>0' The instantaneous state of the junction is specified completely by 8. although higher-order corrections might provide a better fit at voltages above.3) and (3. however. by using . and s at the time e. We should like to thank Dr. Department of Energy. 9. and can also be used to compute the spectral density of the voltage noise. Van HARLINGEN. This work was supported by the Director. 1. and C that yield an excellent fit to the measured 1 and dVld1 vs V characteristics. i=I/lo. is described by Eqs. Once the derivatives have been evaluated numerically for the existing values of B. for reasons that we will explain. Unfortunately. H. D. for example.S.V CHARACTERISTICS AND NOISE IN RESISTIVELY SHUNTED JUNCTIONS In this appendix. We have neglected all derivatives of vn. Greiner and Dr. We rewrite these equations in dimensionless units V=V/loR. This limitation of the Josephson mixer with an external local oscillator has been discussed extensively by other authors. ACKNOWLEDGMENTS R. (A1) and (A2) one can compute sand B and the higher-order derivatives. APPENDIX: TWO METHODS OF COMPUTING 1. Using Eqs. W-7405ENG-48..F which is applicable to a quite different situation. and we have used 2e V =1iaB/at.V characteristics and RD. The first method calculates the 1. we note that the values of L. AND CLARKE 26 of the theory in the noise-rounded case 1 <10 since quantum effects are negligible in this regime in the 4He temperature range for the parameters of our junctions. s=ls/10' and e=t/(<I>oI21710R): (AI) and (A2) where B::=atVae. We emphasize. thereby greatly increasing the noise of the mixer.24. including noise-rounding. e + T.V characteristics and spectral density of the voltage noise for resistively shunted junctions. in accord with other observations. and C to compare the experimental results with the theory. For example.23 Finally. under Contract No. The data from junction 4.4). With the model of the junction we have used.2X 10-22 A2Hz-t. the additional nonlinearity induced on the 1. As usual. 11). However. we outline two methods of computing the 1. etc. Method 1 The model circuit. We are indebted to the Micro Electronics Facility in the Electronics Research Laboratory of the Electrical Engineering and Computer Science Department at the University of California at Berkeley for the use of their facilities. this this observation in no way invalidates the theory of Stephen.84 KOCH. in Fig. The second method calculates the noise very efficiently at voltages where noise rounding is negligible. such as junction 4. Because quantum effects increase rapidly as the order of the harmonic increases. 4. this method is not useful for computing the noise in a junction with resonant structure. Materials Sciences Division of the U. although the last calculation is rather slow. require a fitting of L. 500 fLV. The fact that the zero-point fluctuations in the resistor can be observed at frequencies as high as 5 X 1011 Hz implies that a Josephson mixer using the ac Josephson effect as the local oscillator is an ideal quantum-limited device at these frequencies.V characteristic causes noise near the higher harmonics of the Josephson frequency to be mixed-down. (3. the role of zero-point fluctuations is even more pronounced in junctions with resonant structure. inset in Fig.22. the spectral density of a term 4e10would be about 3. L. also produce a very good fit to the noise data (Fig. Office of Basic Energy Sciences. and s. B. These results show very dramatically the strong effects of additional nonlinearities on the voltage noise due to the mixingdown of higher-order harmonics. When an external local oscillator is used. J.25 we find no evidence for a contribution to the measured noise arising from the shot noise of pairs tunneling through the junction. Koch and D. H. we compute the new values BI> BI> and slat a later time.

was typically 5 to 20 times higher than Wj. To predict the average voltage for i » 1. wLc=(LsCJ-1/2.V characteristic. we find (A6) and (A7) We Fourier-transform these equations over the range . since WL was typically an order of magnitude less than Wj. we had to abandon this technique for junctions with resonant structure.ratio of the "input" noise power to the "output" noise power for "/ noise" was typically 106• The computed spectral densities of the noise proved to be erratic with such large ratios.00 < W < 00 to obtain 0= f_'' 'oo F(w')S(w-w')dw' +(3c( -(2)S(w)+s(w) (A8) and jwS(W)=(1+jW{3L)S(W)+Un(W). following the perturbation approach of Likharev and Semenov. S:'(w) (w is the dimensionless angular frequency). A nonwhite power spectrum S~W(w) could be generated. However. integrate Eqs. by convolving this time representation with an appropriate filter function.lwj. 9. the. 85 a fifth-order Taylor expansion. (Al ) and (A2) numerically over exactly one Josephson cycle. we used a more accurate but more complicated method to calculate the noise in resonant junctions. the computer values of &(()) were filtered with a low-pass Gaussian filter with a roll-off frequency WL of 0. Since cos80(()) is a periodic function. F(w') consists of a series of spiked functions centered at w=o and spaced at intervals of Wj. To obtain a time representation of un(()) with a white power spectrum. while WH was necessarily at least several times greater than WLC' Thus. This method was used in two earlier papersl5. Method 2 Above the noise-rounded region of the J. with values of L. The results were independent of the length of the time step 7 provided 7 was less than the smaller of {3c or (3L' To check that the presence of noise did not affect the characteristics for i > 1. Substituting these expressions into Eqs. the ratio of WH/WL was typically 103• Consequently.26 MEASUREMENTS OF QUANTUM NOISE IN RESISTIVELY . which has substantial resonant struc- ture. measure the required time ()j. we set Un =0. This filter function was chosen so that its transfer function in the frequency domain T (r») satisfied (A3) The high-frequency cutoff. The essential problem was that the resonant frequency.. we obtained very poor results. and Sand srepresent small departures from 80 and So due to noise. The resulting values of the voltage were identical to those with Un =0. where k is an integer. we can . when necessary. In this region. (A9) where F(w') is the normalized Fourier transform of cos80(()). The fluctuations in the filtered values of (&) were used to compute the low-frequency spectral density of the voltage noise. (Al ) and (A2). and C chosen to fit the data.V characteristics where there is significant noise-rounding. WH. (A5) where 80 and So are the noise-free solutions for the phase and shunt current.26 to predict (&) and the fluctuations in & in single junctions and de superconducting quantum interference devices (SQUID's). possibly because of our neglect of the derivatives of Un' As a result. we can expand 8 and s: 8 (A4) and S (()) =so(())+s(()) .. we used a pseudo-randomnumber generator to produce voltage pulses that were Gaussian-distributed in amplitude and uncorrelated in time. including low-voltage regions of the J. when the noise near the Josephson frequency was nonwhite.)=21r/()j=wj. To obtain (&). of un(()) was always chosen to be large enough that the predicted average voltage and noise voltage were independent of the value of WH when the latter was varied over a factor of 20 or more. when we tried to use this method to predict the noise in junction 4. Furthermore. we took account of the implied nonzero correlation time by ensuring that the correlation time of the filter was much larger than lIVj. 2. This spectral density was independent of the roll-off frequency of this low-pass filter. and compute (u)=(f.03 to O. This procedure was used to compute the values of RD in Fig. Setting w'=kwj. we varied Un in time to simulate noise from the resistor while the junction was allowed to evolve over many Josephson cycles.

Scalapino. . 12. is the matrix representation of Eq.B({f) + Im{f)J). Phys. Rogovin.k [ 1 +j({f)+lwJ ){3L -(W+IWJ)2{3c I. Josephson. Phys. 42. Phys. B 10. J. We find +km B({f)+lwJ)= k=-km l: where 4 =ll. (All) with 1=0: sil(w)= where W1 « 1 WJ 1 and i FJJ({f)-k{f)J)+ [1 +l j:{3 k=-oo L _(f)2{3c IB({f)) (AlO) k=-km l: km IZ k(W)12Sj({f)+kwJ). Phys. Lett. and eliminate S between Eqs. Van HARLINGEN. Stewart. 14R. Soulen. In dimensioned units. 31A. 22. J. "Present address: Department of Physics and Materials Research Laboratory. 9R. V. and Sj(w + kWJ) is the dimensionless spectral density of the noise current in the resistor..442 (1972)]. F. tiJ· z. IW. Giffard. Rev. replace F(k{f)J) with Fk. -I. Appl. The required fluctuations in the B({f)-I{f)J) are then obtained by a conventional matrix inversion of Eq. Phys. H. 182. K. K. 32. The computed values of Z (w) were shown to be independent of W for W «Wb and w/{f)J was chosen to be between 3~ and The value of km' typically 16 to 25. Lett. Rev. 314 (1970). and R. C. (A 13) (AI4) In subsequent calculations. Ambegoakar. llR. 12. A.625 (1972) [JETP Lett. Gubankov. P. and J.251 (1962). McCumber. D. (AlO): !ll. Koch. Rev. Denenstein. 11. Appl. K.86 KOCH. 3113 (1968).k is the Kronecker delta. 79 (1974).. we have shown that there exists a maximum value of 1 k 1 . using positive frequencies only. (A12) In Eq. was chosen so that km{f)J »(f)L6 the value of km was varied to show that the values of Z(w) did not depend on it. Rev. Koch and J. and D. The method was used to compute the spectral density of junction 4 shown in Fig. (AI3) by replacing (f) with v. Jr. (3. 11 was computed with R =0. the noise at the measurement frequency can be ob- is the complex dimensionless impedance that mixes noise from (f) + k{f)J to (f) + Iw. Urbana. AND CLARKE 26 1 transform the integral to a sum. 8K. N. K. V. (A8) and (A9) to find tained from Eq. Lett. J. Since the vn at different frequenices are independent. Semenov. N. Kuzmin. Ambegoakar and B. IL 61801. H. 22. km' above which the noise at frequency koi. L. Kurkijarvi and V. Hansma. W. Cutting off the summation at ±km' we are left with 2km + 1 inhomogeneous equations with unknown phases B({f)-lm{f)J). J. Phys. 12. Lett. Phys. Phys. Phys. 7R. J. 12M. and the corresponding values of 1 12 in Fig. 6H. and ll. and the noise in Fig. 2D.. Fiz. setting v=O.. Van Harlingen. This approximation gave rise to the discrepancy between the measured and predicted noise at low voltages in Fig. 1416 (1969). E. Appl.k =Ft-k j({f)+I{f)J) +cSt. Phys. Likharev. 13A. Halperin. N. Phys. K. D. is not significantly mixed-down to the measurement frequency. E. Phys. Rev. 1. J. where 1 I 1 :::. Clarke (unpublished). V. 12. Zo is just the dynamic resistance. Teor. Falco. Parker. Rev. at frequencies small compared with Zo/Ls. Trullinger. 3B.}t. Langenberg. Vystavkin. Parker. Lett. Pis'ma Red. K. Semenov. S. Low Temp. 45. 1364 (1969). the method can be tested by comparing the value of Zo with the value of RD obtained with method 1. 2. we first compute the coefficients Fk using method 1 (with Vn =0). km• To solve these. D.. 151 (1981). . and P. (AlO). H. Dahm. Lett. 39. . S. I. 277 (1968). Stephen. Thus. Appl. M. 2132 (1980). Zh. 4V. Clarke.770 (1978).531 (1969). 11. Voss. Likharev and V. Lett. and V. 5A. and assigning appropriate dimensions. D. The complexity of the method does not easily allow the value of R to be voltage dependent. IOC.7) from Eq. W. 8/. H. Migulin.092 n for all voltages. We obtain Eq. Eksp. J. (AI2). 1865 (1974). University of Illinois at Urbana-Champaign.

18J. Ivanchenko and L. Eksp. R. 21J. P. J.26 ISH. Taur. Michelson.H. J. 272 (1969)]. Callen and T. 211 (1981). Baker. 55. Y. 34 (1951). 22J. Yen. 38. P. Giffard. Phys. Lett. H. in Superconducting Quantum Interference Devices and their Applications.247. W. D. T. Klepner. ~. Claassen and P. Warnecke. J. edited by H. Fiz.2395 (1968) [Sov. 87 B. Dev. p. E. 759 (1974). Van Harlingen. 2826 (1976).o. Soulen. Greiner. K. Zh. Mu- rakami. S.-C. Res. Clarke. Phys. J.. Claassen. P.Kurkijiirvi. Koch. 1921 (1980). 26R.!U. A. H. Liibbig (de Gruyter. L. 380 (1981). 20J.Clarke and G. Hawkins. Lett. Phys. Brosious. Berlin. Basavaiah. Appl. L. and I. MEASUREMENTS OF QUANTUM NOISE IN RESISTIVELY .H. Teor. 276 (1979). 24. IBM J. Phys. . and P. 4117 (1978). Leggett. Leggett. B 14. P. MAG-15. J. J.. Appl. 16yu.J. IEEE Trans.-JETP 28. Phys. Phys. and J. H. Lahiri. Caldeira and A. Zil'berman. Taur. 2sR. 17A. F. A. IEEE Trans. 23y. M. 195 (1980). Phys. Ames. . Rev. (Paris)~. 25. Appl. John M. Electron Devices ED-27. 19A. Rev. Magn. 24J. Hahlbohm and H. Lett. M. C. S. Welton. Huang. Richards. Rev. J. 46. and R. Richards. S. A. D. Kircher. 1264 (1978). 1980). Phys.