betaFeedback

WorksPhysics and Astronomy

Multiphysics Numerical Method for Modeling Josephson Traveling-Wave Parametric Amplifiers

Abstract

Josephson traveling-wave parametric amplifiers (JTWPAs) are wideband, ultralow-noise amplifiers used to enable the readout of superconducting qubits. While individual JTWPAs have achieved high performance, behavior between devices is inconsistent due to wide manufacturing tolerances. Amplifier designs could be modified to improve resilience towards variations in amplifier components; however, existing device models often rely on analytical techniques that typically fail to incorporate component variations. To begin addressing this issue, a 1D numerical method for modeling JTWPAs is introduced in this work. The method treats the Josephson junctions and transmission lines in an amplifier as coupled subsystems and can easily incorporate arbitrary parameter variations. We discretize the transmission line subsystem with a finite element time domain method and the Josephson junction subsystem with a finite difference method, with leapfrog time marching used to evolve the system in time. We validate our method by comparing the computed gain to an analytical model for a traditional JTWPA architecture and one with resonant phase matching. We then use our method to demonstrate the impact of variations in Josephson junctions and phase-matching resonators on amplification. In future work, the method will be adjusted to incorporate additional amplifier architectures and extended to a 3D full-wave approach.

Introduction

Superconducting qubits are a leading architecture in the race towards large-scale quantum computing [1, 2, 3]. However, performance improvements are still needed in many of the system components, including those used to measure the states of qubits. To measure a qubit state, an extremely low-power microwave pulse is used to measure the scattering properties of an electromagnetic resonator coupled to the qubit [4, 5]. To perform high-fidelity measurements of the weak microwave signals that carry information about the qubit state, an exceptionally low-noise amplifier (i.e., close to the standard quantum limit) with high gain is required [6, 7].

Originally investigated in the late 1960’s [8, 9], interest in Josephson parametric amplifiers has been renewed due to their potential for meeting these needs of superconducting qubit technologies [10]. These devices use a superconducting Josephson junction to facilitate a nonlinear mixing operation with a strong pump, producing parametric amplification of the desired signal. To improve the achievable gain, the signal is coupled into the nonlinear element through a resonator for high interaction time; however, this comes at the cost of a constrained amplifier bandwidth [7]. Josephson traveling-wave parametric amplifiers (JTWPAs) were proposed as a wideband alternative, but their lackluster gain combined with challenging manufacturing requirements made them difficult to implement [11]. While these devices can achieve long interaction times by embedding thousands of Josephson junctions in a transmission line, amplification is inhibited by phase mismatches produced through pump-induced dispersion [12]. However, modern fabrication techniques have enabled the development of resonant phase matching (RPM), where resonant loads are periodically embedded into the transmission line to compensate for deleterious phase mismatch and obtain significantly higher gain [12, 13]. Over time, the RPM technique has led JTWPAs to become a promising technology for enabling qubit readout due to their wide bandwidth, high gain, and ultra low-noise operation [11, 14, 15, 16]. However, alternative techniques for achieving phase matching have also been developed, such as periodically varying the transmission line impedance [17] and modifying the nonlinearity through flux biasing of Josephson junction-based circuit elements [18].

While modern fabrication technologies and design techniques have allowed high performance JTWPAs to be realized, their broader application is still limited by manufacturing variations. Fluctuations between junction properties across a device can produce impedance mismatches, leading to inconsistent performance between JTWPAs [19, 20, 21]. Additionally, even small variations in the resonant loads of RPM architectures can disrupt the delicate phase matching relationship and significantly reduce gain [22]. Ideally, JTWPA designers could readily simulate the impact of these variations during the design process and make adjustments to reduce their severity. However, most JTWPA models treat the device as a continuum [17, 23, 24, 25] to develop analytical solutions, preventing variations between elements of the circuit from being modeled.

To combat this issue, previous approaches have accounted for parameter variations by introducing a position-dependent wavenumber [12, 22, 26]. In these methods, analytical expressions are used to describe segments of a JTWPA with a fixed wavenumber, which are stitched together numerically to simulate the whole JTWPA. While these approaches can accommodate phase mismatch due to parameter variations, they do not incorporate impedance mismatches between segments, which cause reflections that alter performance. As a result, these methods cannot fully capture the impact that manufacturing variability can have on a JTWPA design. Aspects of these effects can be captured in numerical methods using

Manuscript received XX XX, 2024; revised XX XX, 2024.

Samuel T. Elkin and Thomas E. Roth are with the Elmore Family School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907 USA and the Purdue Quantum Science and Engineering Institute, West Lafayette, IN 47907 USA (e-mail: [email protected])

Michael Haider is with the TUM School of Computation, Information and Technology, Technical University of Munich, 85748 Garching, Germany lumped-element circuit simulators such as WRSPICE [27, 28, 29]. However, these approaches, along with the others, are not naturally extensible to a 3D full-wave method, preventing them from modeling explicit device geometries. Instead, these methods must estimate circuit parameters from the explicit geometry, which can often be a user-intensive and potentially inaccurate procedure that is avoided in a full-wave approach.

In this work, we develop a 1D multiphysics numerical method which treats the Josephson junctions and transmission lines in a JTWPA as coupled dynamical systems. The transmission lines are discretized via a finite element time domain (FETD) approach, while the differential equations describing junction dynamics are discretized using finite differencing. Interaction between the dynamical variables is facilitated through a leap-frog time-marching procedure. In this treatment, individual JTWPA elements can easily be varied to model arbitrary distributions for the properties of those components. Additionally, this method inherently considers higher-order nonlinear effects such as the generation and impact of harmonics. As JTWPA designs evolve, understanding these effects is increasingly important, but they are difficult to characterize using analytical approaches [27]. Similarly, pump depletion is naturally accounted for in this formulation, allowing the dynamic range of a design to easily be determined. The method can also be extended to a 3D full-wave description in the future to provide superior accuracy in addition to the aforementioned benefits.

Preliminary results on this formulation were reported in [30]. This work significantly expands on [30] by providing a more detailed formulation of the dynamical equations, as well as extending the method to incorporate RPM. Further, complete details on the numerical discretization approach are also presented. Finally, new numerical results are presented, including a case-study analyzing the impact of process variations in resonant loads and Josephson junctions.

This work is structured as follows. In § 2, the equations of motion and boundary conditions for a unit cell of a JTWPA are derived, both with and without RPM. Following that, Section III develops the 1D hybrid method to discretize a unit cell, and extends the description to a full JTWPA for each case. Section IV validates the method using analytical results for each type of architecture, and demonstrates the effects of variations in the properties of Josephson junctions and resonant loads on amplification. In § 5, we discuss the conclusions drawn from these results and the direction for future development of the method.

Formulation

In this section, the equations of motion (EoMs) and boundary conditions applicable to a single JTWPA unit cell are derived using techniques from the Lagrangian mechanics of continuous systems [31]. In Section II-A, this is done for the “basic JTWPA unit cell” of Figure 1, while in Section II-B the derivation is extended to include a resonant load so architectures using RPM can be analyzed.

Image not extracted

Figure 1. Schematic of a basic single JTWPA unit cell, consisting of a Josephson junction with a nonlinear inductance (LJL_J, denoted by an “X”) in parallel with a linear capacitance (CJC_J).

Basic JTWPA Unit Cell

Deriving EoMs and boundary conditions for the unit cell of Figure 1 using Lagrangian mechanics requires a Lagrangian for each portion of the system. Starting with the transmission lines, the Lagrangian for each line can be written as

LTXi=12∫(CiV2(z,t)−LiI2(z,t)) dz,(1)\mathcal{L}_{\mathrm{TX}i} = \frac{1}{2} \int\left(C_i V^2(z,t) - L_i I^2(z,t)\right)\,\mathrm{d}z, \tag*{(1)}

where CiC_i and LiL_i are the per-unit-length capacitance and inductance of the transmission line. However, to derive EoMs the Lagrangian must be expressed in terms of a generalized coordinate and its derivatives rather than voltage and current. To accommodate this requirement, we use the node flux ϕ(z,t)\phi(z,t), which is related to the voltage and current by ϕ˙(z,t)=V(z,t)\dot{\phi}(z,t) = V(z,t) and ϕ′(z,t)=−LI(z,t)\phi'(z,t) = -L I(z,t), where ϕ˙(z,t)\dot{\phi}(z,t) and ϕ′(z,t)≡∂∂zϕ(z,t)\phi'(z,t) \equiv\frac{\partial}{\partial z}\phi(z,t) [32]. The resulting Lagrangian is

LTXi=12∫(Ciϕ˙2(z,t)−Li−1ϕ′2(z,t)) dz.(2)\mathcal{L}_{\mathrm{TX}i} = \frac{1}{2} \int\left(C_i\dot{\phi}^{2}(z,t) - L_i^{-1}\phi'^{2}(z,t)\right)\,\mathrm{d}z. \tag*{(2)}

The Josephson junction Lagrangian is developed in [33] as

LJJ=CJ2ϕ˙J2(t)+ℏIc2ecos⁡(2eℏϕJ(t)),(3)\mathcal{L}_{\mathrm{JJ}} = \frac{C_J}{2}\dot{\phi}_J^2(t) + \frac{\hbar I_c}{2e}\cos\left(\frac{2e}{\hbar}\phi_J(t)\right), \tag*{(3)}

where ee is the elementary charge, ℏ\hbar is the reduced Planck’s constant, and ϕJ(t)≡ϕ(z1,t)−ϕ(z2,t)\phi_J(t) \equiv\phi(z_1,t)-\phi(z_2,t) is the junction flux. The Josephson junction is characterized by the junction capacitance CJC_J and critical current IcI_c, which is related to the Josephson inductance LJL_J through Ic=ℏ/(2eLJ)I_c = \hbar/(2eL_J).

To derive the EoMs and boundary conditions in Lagrangian mechanics, we must first introduce the action functional, S=∫L dtS = \int\mathcal{L}\,\mathrm{d}t. The principle of least action states that a physical system will always operate such that the action is minimized, meaning δS=0\delta S = 0 [31]. Typically, the principle of least action can be satisfied using the Euler-Lagrange equation. However, for our continuous systems with non-trivial interconnections as in Figure 1, the typical Euler-Lagrange equation is no longer applicable. Hence, here we instead take the variation of the action functional directly. We simplify this process by decomposing δS\delta S into terms for each subsystem,

δS=δSTX1+δSTX2+δSJJ.(4)\delta S = \delta S_{\mathrm{TX}_1} + \delta S_{\mathrm{TX}_2} + \delta S_{\mathrm{JJ}}. \tag*{(4)}

For the first transmission line, δSTX1\delta S_{\mathrm{TX1}} is given by

δSTX1=∫(δϕ˙(z,t)∂LTX1∂ϕ˙(z,t)+δϕ′(z,t)∂LTX1∂ϕ′(z,t))dt(5)\delta S_{\mathrm{TX1}}=\int\left(\delta\dot{\phi}(z,t)\frac{\partial\mathcal{L}_{\mathrm{TX1}}}{\partial\dot{\phi}(z,t)}+\delta\phi'(z,t)\frac{\partial\mathcal{L}_{\mathrm{TX1}}}{\partial\phi'(z,t)}\right)dt \tag*{(5)}

By evaluating each derivative using (2), (5) becomes

δSTX1=∫∫0z1(C1ϕ˙(z,t)δϕ˙(z,t)−L1−1ϕ′(z,t)δϕ′(z,t))dz dt(6)\delta S_{\mathrm{TX1}}=\int\int_{0}^{z_{1}}\left(C_{1}\dot{\phi}(z,t)\delta\dot{\phi}(z,t)-L_{1}^{-1}\phi'(z,t)\delta\phi'(z,t)\right)dz\,dt \tag*{(6)}

To remove derivatives from the variations, we employ integration by parts. For the temporal variation δϕ˙(z,t)\delta\dot{\phi}(z,t), it is appropriate to consider “fixed endpoint” variations so that the boundary terms from the integration by parts can be neglected [31]. However, for the spatial variation δϕ′(z,t)\delta\phi'(z,t), we need to consider variations with “variable endpoints” to eventually derive a spatial boundary condition at z=z1z=z_{1} that accounts for the effect of the Josephson junction [31]. As a result, (6) becomes

δSTX1=∫(∫0z1(−C1ϕ¨(z,t)+L1−1ϕ′′(z,t))δϕ(z,t) dz−L1−1ϕ′(z1,t)δϕ(z1,t))dt(7)\delta S_{\mathrm{TX1}}=\int\left(\int_{0}^{z_{1}}\left(-C_{1}\ddot{\phi}(z,t)+L_{1}^{-1}\phi''(z,t)\right)\delta\phi(z,t)\,dz-L_{1}^{-1}\phi'(z_{1},t)\delta\phi(z_{1},t)\right)dt \tag*{(7)}

Similarly, for the other transmission line we find δSTX2\delta S_{\mathrm{TX2}} is

δSTX2=∫(∫z2a(−C2ϕ¨(z,t)+L2−1ϕ′′(z,t))δϕ(z,t) dz+L2−1ϕ′(z2,t)δϕ(z2,t))dt(8)\delta S_{\mathrm{TX2}}=\int\left(\int_{z_{2}}^{a}\left(-C_{2}\ddot{\phi}(z,t)+L_{2}^{-1}\phi''(z,t)\right)\delta\phi(z,t)\,dz+L_{2}^{-1}\phi'(z_{2},t)\delta\phi(z_{2},t)\right)dt \tag*{(8)}

Moving to the Josephson junction, the change of independent variables in (3) leads to the expression

δSJJ=∫(δϕJ(t)∂L∂ϕJ(t)+δϕ˙J(t)∂L∂ϕ˙J(t))dt(9)\delta S_{\mathrm{JJ}}=\int\left(\delta\phi_{J}(t)\frac{\partial\mathcal{L}}{\partial\phi_{J}(t)}+\delta\dot{\phi}_{J}(t)\frac{\partial\mathcal{L}}{\partial\dot{\phi}_{J}(t)}\right)dt \tag*{(9)}

By evaluating the derivatives using (3), we obtain

δSJJ=∫[CJϕ˙J(t)(δϕ˙(z1,t)−δϕ˙(z2,t))−Icsin⁡(2eℏϕJ(t))(δϕ(z1,t)−δϕ(z2,t))]dt(10)\delta S_{\mathrm{JJ}}=\int\left[C_{J}\dot{\phi}_{J}(t)\left(\delta\dot{\phi}(z_{1},t)-\delta\dot{\phi}(z_{2},t)\right)-I_{c}\sin\left(\frac{2e}{\hbar}\phi_{J}(t)\right)\left(\delta\phi(z_{1},t)-\delta\phi(z_{2},t)\right)\right]dt \tag*{(10)}

where substitutions were made for δϕJ(t)\delta\phi_{J}(t) and δϕ˙J(t)\delta\dot{\phi}_{J}(t) using ϕJ(t)=ϕ(z1,t)−ϕ(z2,t)\phi_{J}(t)=\phi(z_{1},t)-\phi(z_{2},t). Integrating by parts produces

δSJJ=−∫[CJϕ¨J(t)(δϕ(z1,t)−δϕ(z2,t))+Icsin⁡(2eℏϕJ(t))(δϕ(z1,t)−δϕ(z2,t))]dt(11)\delta S_{\mathrm{JJ}}=-\int\left[C_{J}\ddot{\phi}_{J}(t)\left(\delta\phi(z_{1},t)-\delta\phi(z_{2},t)\right)+I_{c}\sin\left(\frac{2e}{\hbar}\phi_{J}(t)\right)\left(\delta\phi(z_{1},t)-\delta\phi(z_{2},t)\right)\right]dt \tag*{(11)}

which is used alongside (7) and (8) to evaluate (4),

δS=∫[∫0z1(−C1ϕ¨(z,t)+L1−1ϕ′′(z,t))δϕ(z,t) dz−(L1−1ϕ′(z1,t)+CJϕ¨J(t)+Icsin⁡(2eℏϕJ(t)))δϕ(z1,t)+(L2−1ϕ′(z2,t)+CJϕ¨J(t)+Icsin⁡(2eℏϕJ(t)))δϕ(z2,t)+∫z2a(−C2ϕ¨(z,t)+L2−1ϕ′′(z,t))δϕ(z,t) dz]dt(12)\delta S=\int\left[\int_{0}^{z_{1}}\left(-C_{1}\ddot{\phi}(z,t)+L_{1}^{-1}\phi''(z,t)\right)\delta\phi(z,t)\,dz-\left(L_{1}^{-1}\phi'(z_{1},t)+C_{J}\ddot{\phi}_{J}(t)+I_{c}\sin\left(\frac{2e}{\hbar}\phi_{J}(t)\right)\right)\delta\phi(z_{1},t)+\left(L_{2}^{-1}\phi'(z_{2},t)+C_{J}\ddot{\phi}_{J}(t)+I_{c}\sin\left(\frac{2e}{\hbar}\phi_{J}(t)\right)\right)\delta\phi(z_{2},t)+\int_{z_{2}}^{a}\left(-C_{2}\ddot{\phi}(z,t)+L_{2}^{-1}\phi''(z,t)\right)\delta\phi(z,t)\,dz\right]dt \tag*{(12)}

The principle of least action states that δS=0\delta S=0 for all tt. Since each variation is nonzero by definition, the coefficient of each variation in (12) must equate to zero. This step produces the equations of motion and boundary conditions,

ϕ′′(z,t)−L1C1ϕ¨(z,t)=0,0<z<z1,(13)\phi''(z,t)-L_{1}C_{1}\ddot{\phi}(z,t)=0,\qquad0<z<z_{1}, \tag*{(13)}
L1−1ϕ′(z1,t)=−CJϕ¨J(t)−Icsin⁡(2eℏϕJ(t)),(14)L_{1}^{-1}\phi'(z_{1},t)=-C_{J}\ddot{\phi}_{J}(t)-I_{c}\sin\left(\frac{2e}{\hbar}\phi_{J}(t)\right), \tag*{(14)}
L2−1ϕ′(z2,t)=−CJϕ¨J(t)−Icsin⁡(2eℏϕJ(t)),(15)L_{2}^{-1}\phi'(z_{2},t)=-C_{J}\ddot{\phi}_{J}(t)-I_{c}\sin\left(\frac{2e}{\hbar}\phi_{J}(t)\right), \tag*{(15)}
ϕ′′(z,t)−L2C2ϕ¨(z,t)=0,z2<z<a,(16)\phi''(z,t)-L_{2}C_{2}\ddot{\phi}(z,t)=0,\qquad z_{2}<z<a, \tag*{(16)}

Clearly, (13) and (16) describe waves propagating down each transmission line. The boundary conditions (14) and (15) introduce nonlinearity through interactions with the Josephson junction. To move towards allowing this nonlinear set of equations to be solved with a linear time-marching procedure, we treat ϕJ(t)\phi_{J}(t) as a second dynamical variable with EoM

CJϕ¨J(t)+Icsin⁡(2eℏϕJ(t))=−12L1ϕ′(z1,t)−12L2ϕ′(z2,t).(17)C_{J}\ddot{\phi}_{J}(t)+I_{c}\sin\left(\frac{2e}{\hbar}\phi_{J}(t)\right)=-\frac{1}{2L_{1}}\phi'(z_{1},t)-\frac{1}{2L_{2}}\phi'(z_{2},t). \tag*{(17)}

This is obtained by summing (14) and (15). A numerical procedure to enable a linear solution for time-marching the dynamical variables in (13), (16), and (17) together will be discussed in Section III.

RPM JTWPA Unit Cell

To implement RPM, resonant loads of capacitance CrC_{r} and inductance LrL_{r} are periodically coupled into unit cells through capacitance CcC_{c}, as shown in Figure 2. To determine how this change impacts the equations of motion, we reconstruct the Lagrangian for the second transmission line, which becomes

Schematic of a JTWPA unit cell for a RPM design

Figure 2. Schematic of a JTWPA unit cell for a RPM design. The added resonant load and coupling capacitor are outlined in red.

LTX2=12∫z2a(C2ϕ˙2(z,t)−L2−1ϕ′2(z,t))dz+12Cc(ϕ˙(z3,t)−ϕ˙r(t))2+12(Crϕ˙r2(t)−Lr−1ϕr2(t)).(18)\mathcal{L}_{\mathrm{TX2}}=\frac{1}{2}\int_{z_{2}}^{a}\left(C_{2}\dot{\phi}^{2}(z,t)-L_{2}^{-1}\phi'^{2}(z,t)\right)dz+\frac{1}{2}C_{c}\left(\dot{\phi}(z_{3},t)-\dot{\phi}_{r}(t)\right)^{2}+\frac{1}{2}\left(C_{r}\dot{\phi}_{r}^{2}(t)-L_{r}^{-1}\phi_{r}^{2}(t)\right). \tag*{(18)}

where ϕr(t)\phi_r(t) is the flux stored in the resonator. Therefore, the variation of the action functional is given by

δSTX2=∫(δϕr(z,t)∂L∂ϕr(z,t)+δϕ˙r(z,t)∂L∂ϕ˙r(z,t)+δϕ′(z,t)∂L∂ϕ′(z,t)+δϕ˙(z,t)∂L∂ϕ˙(z,t)+δϕ˙(z3,t)∂L∂ϕ˙(z3,t)) dt.(19)\delta S_{\mathrm{TX}_2}=\int\left(\delta\phi_r(z,t)\frac{\partial\mathcal{L}}{\partial\phi_r(z,t)}+\delta\dot{\phi}_r(z,t)\frac{\partial\mathcal{L}}{\partial\dot{\phi}_r(z,t)}+\delta\phi'(z,t)\frac{\partial\mathcal{L}}{\partial\phi'(z,t)}+\delta\dot{\phi}(z,t)\frac{\partial\mathcal{L}}{\partial\dot{\phi}(z,t)}+\delta\dot{\phi}(z_3,t)\frac{\partial\mathcal{L}}{\partial\dot{\phi}(z_3,t)}\right)\,dt. \tag*{(19)}

Substituting (18) into (19) produces the expression

δSTX2=∫(∫z2a(C1ϕ˙(z,t)δϕ˙(z,t)−L1−1ϕ′(z,t)δϕ′(z,t)) dz+Crϕ˙r(t)δϕ˙r(t)−Lr−1ϕr(t)δϕr(t)+Cc(ϕ˙(z3,t)−ϕ˙r(t))(δϕ˙(z3,t)−δϕ˙r(t))) dt,(20)\delta S_{\mathrm{TX}_2}=\int\left(\int_{z_2}^{a}\left(C_1\dot{\phi}(z,t)\delta\dot{\phi}(z,t)-L_1^{-1}\phi'(z,t)\delta\phi'(z,t)\right)\,dz+C_r\dot{\phi}_r(t)\delta\dot{\phi}_r(t)-L_r^{-1}\phi_r(t)\delta\phi_r(t)+C_c\left(\dot{\phi}(z_3,t)-\dot{\phi}_r(t)\right)\left(\delta\dot{\phi}(z_3,t)-\delta\dot{\phi}_r(t)\right)\right)\,dt, \tag*{(20)}

which after integrating by parts becomes

δSTX2=∫[∫z2a(−C2ϕ¨(z,t)+L2−1ϕ′′(z,t))δϕ(z,t) dz+L2−1ϕ′(z2,t)δϕ(z2,t)−(Crϕ¨r(t)−Lr−1ϕr(t)+Cc(ϕ¨(z3,t)−ϕ¨r(t)))δϕr(t)−Cc(ϕ¨(z3,t)−ϕ¨r(t))δϕ(z3,t)] dt.(21)\delta S_{\mathrm{TX}_2}=\int\left[\int_{z_2}^{a}\left(-C_2\ddot{\phi}(z,t)+L_2^{-1}\phi''(z,t)\right)\delta\phi(z,t)\,dz+L_2^{-1}\phi'(z_2,t)\delta\phi(z_2,t)-\left(C_r\ddot{\phi}_r(t)-L_r^{-1}\phi_r(t)+C_c\left(\ddot{\phi}(z_3,t)-\ddot{\phi}_r(t)\right)\right)\delta\phi_r(t)-C_c\left(\ddot{\phi}(z_3,t)-\ddot{\phi}_r(t)\right)\delta\phi(z_3,t)\right]\,dt. \tag*{(21)}

This expression can be used in conjunction with (7) and (11) to determine δS\delta S for the whole RPM unit cell,

\begin{aligned} \delta S={}&\int\left[\int_{0}^{z_1}\left(-C_1\ddot{\phi}(z,t)+L_1^{-1}\phi''(z,t)\right)\delta\phi(z,t)\,dz\\ &+\int_{z_2}^{a}\left(-C_2\ddot{\phi}(z,t)+L_2^{-1}\phi''(z,t)\right)\delta\phi(z,t)\,dz\\ &-\left(L_1^{-1}\phi'(z_1,t)+\frac{C_J}{2}\ddot{\phi}_J(t)+I_c\sin\left(\frac{2e}{\hbar}\phi_J(t)\right)\right)\delta\phi(z_1,t)\\ &+\left(L_2^{-1}\phi'(z_2,t)+\frac{C_J}{2}\ddot{\phi}_J(t)+I_c\sin\left(\frac{2e}{\hbar}\phi_J(t)\right)\right)\delta\phi(z_2,t)\\ &-\left(C_r\ddot{\phi}_r(t)-L_r^{-1}\phi_r(t)+C_c\left(\ddot{\phi}(z_3,t)-\ddot{\phi}_r(t)\right)\right)\delta\phi_r(t)\\ &-C_c\left(\ddot{\phi}(z_3,t)-\ddot{\phi}_r(t)\right)\delta\phi(z_3,t)\right]\,dt. \tag*{(22)} \end{aligned}

Applying the principle of least action to (22) leads to

ϕ′′(z,t)−L1C1ϕ¨(z,t)=0,0<z<z1,(23)\phi''(z,t)-L_1C_1\ddot{\phi}(z,t)=0,\quad0<z<z_1, \tag*{(23)}
L1−1ϕ′(z1,t)=−CJϕ¨J(t)−Icsin⁡(2eℏϕJ(t)),(24)L_1^{-1}\phi'(z_1,t)=-C_J\ddot{\phi}_J(t)-I_c\sin\left(\frac{2e}{\hbar}\phi_J(t)\right), \tag*{(24)}
L2−1ϕ′(z2,t)=−CJϕ¨J(t)−Icsin⁡(2eℏϕJ(t)),(25)L_2^{-1}\phi'(z_2,t)=-C_J\ddot{\phi}_J(t)-I_c\sin\left(\frac{2e}{\hbar}\phi_J(t)\right), \tag*{(25)}
ϕ′′(z,t)−L2C2ϕ¨(z,t)=0,z2<z<z3,(26)\phi''(z,t)-L_2C_2\ddot{\phi}(z,t)=0,\quad z_2<z<z_3, \tag*{(26)}
ϕ′′(z3,t)−L2(C2+Cc)ϕ¨(z3,t)=−L2Ccϕ¨r(t),(27)\phi''(z_3,t)-L_2(C_2+C_c)\ddot{\phi}(z_3,t)=-L_2C_c\ddot{\phi}_r(t), \tag*{(27)}
ϕ′′(z,t)−L2C2ϕ¨(z,t)=0,z3<z<a,(28)\phi''(z,t)-L_2C_2\ddot{\phi}(z,t)=0,\quad z_3<z<a, \tag*{(28)}
ϕr(t)−Lr(Cr+Cc)ϕ¨r(t)=−LrCcϕ¨(z3,t).(29)\phi_r(t)-L_r(C_r+C_c)\ddot{\phi}_r(t)=-L_rC_c\ddot{\phi}(z_3,t). \tag*{(29)}

Since (24) and (25) are unchanged from (14) and (15), (17) still holds for the RPM unit cell. A numerical procedure to enable a linear solution for time-marching the dynamical variables in (17), (23), and (26) to (29) together will be discussed in Section III.

Discretization

Here, we detail the development of the hybrid model from the EoMs for each unit cell. In this method, 1D FETD is applied to discretize ϕ(z,t)\phi(z,t) along the transmission lines. FETD is chosen to simplify the transition to a 3D full-wave implementation in the future. Temporal discretization of ϕJ(t)\phi_J(t) is performed using central differencing, which will be demonstrated to produce a linear time-marching equation. The direct relationship between the dynamical variables is eliminated using leap-frog time-marching, which allows ϕ(z,t)\phi(z,t) to be evaluated without the need for a nonlinear solver.

This section is structured as follows. First, the method for the basic JTWPA unit cell is developed in Section III-A by deriving time-marching equations for the dynamical variables. Additionally, we introduce the leap-frog time-stepping procedure to linearly advance the dynamical variables in time. In Section III-B, the unit cells are combined and integrated with driving and terminating components to model a full JTWPA. Finally, the time-marching equations are adjusted to accommodate the addition of resonant loads to model a RPM JTWPA in Section III-C.

Basic JTWPA Unit Cell

To begin, we discretize the node flux ϕ(z,t)\phi(z,t) along the transmission lines, which is described by (13) to (16). We follow a standard FETD process outlined in [32, 34] with first-order (triangular) nodal basis functions Ni(z)N_i(z) such that

ϕ(z,t)=∑iϕi(t)Ni(z).(30)\phi(z,t)=\sum_i\phi_i(t)N_i(z). \tag*{(30)}

This process leads to a matrix equation of the form

[Tu]d2dt2{ϕ(t)}+[Su]{ϕ(t)}={fu(t)}.(31)[T_u]\frac{d^2}{dt^2}\{\phi(t)\}+[S_u]\{\phi(t)\}=\{f_u(t)\}. \tag*{(31)}

where {ϕ(t)}i=ϕi(t)\{\phi(t)\}_i=\phi_i(t) and

[Tu]ij=L1C1∫0z1Ni(z)Nj(z) dz+L1CJ(δjM−δj(M+1))δiM+L2C2∫z2aNi(z)Nj(z) dz−L2CJ(δjM−δj(M+1))δi(M+1),(32)\begin{aligned} [T_u]_{ij}={}&L_1C_1\int_0^{z_1}N_i(z)N_j(z)\,dz\\ &+L_1C_J(\delta_{jM}-\delta_{j(M+1)})\delta_{iM}+L_2C_2\int_{z_2}^{a}N_i(z)N_j(z)\,dz\\ &-L_2C_J(\delta_{jM}-\delta_{j(M+1)})\delta_{i(M+1)}, \tag*{(32)} \end{aligned}
[Su]ij=∫0z1Ni′(z)Nj′(z) dz+∫z2aNi′(z)Nj′(z) dz,(33)[S_u]_{ij}=\int_0^{z_1}N_i'(z)N_j'(z)\,dz+\int_{z_2}^{a}N_i'(z)N_j'(z)\,dz, \tag*{(33)}
{fu(t)}i=Ic(L2δi(M+1)−L1δiM)sin⁡(2eℏϕJ(t)).(34)\{f_u(t)\}_i=I_c(L_2\delta_{i(M+1)}-L_1\delta_{iM})\sin\left(\frac{2e}{\hbar}\phi_J(t)\right). \tag*{(34)}

where δij\delta_{ij} is a Kronecker delta, and nodes MM and M+1M+1 are positioned at z1z_1 and z2z_2 respectively. For the junction flux, applying (30) leads (17) to become

CJϕ¨J(t)+Icsin⁡(2eℏϕJ(t))=∑i(−12L1Ni′(z1)−12L2Ni′(z2))ϕi(t)=−12L1Δz(ϕM(t)−ϕM−1(t))−12L2Δz(ϕM+2(t)−ϕM+1(t)),(35)\begin{aligned} C_J\ddot{\phi}_J(t)+I_c\sin\left(\frac{2e}{\hbar}\phi_J(t)\right)=\sum_i\left(-\frac{1}{2L_1}N_i'(z_1)-\frac{1}{2L_2}N_i'(z_2)\right)\phi_i(t) \\ &=-\frac{1}{2L_1\Delta z}\left(\phi_M(t)-\phi_{M-1}(t)\right)-\frac{1}{2L_2\Delta z}\left(\phi_{M+2}(t)-\phi_{M+1}(t)\right), \tag*{(35)} \end{aligned}

where the final equality was reached by taking piecewise derivatives of the triangular basis functions.

At this point, {ϕ(t)}\{\phi(t)\} and ϕJ(t)\phi_J(t) are still directly related through ϕJ(t)≡ϕM(t)−ϕM+1(t)\phi_J(t)\equiv\phi_M(t)-\phi_{M+1}(t), which invalidates the source treatment of ϕJ(t)\phi_J(t) in (34). However, leap-frogging between the dynamical variables in the time discretization can be exploited to circumvent this issue. More explicitly, by solving for each dynamical variable at alternating half-timesteps, the direct relationship between them is no longer enforced, enabling the “source treatment” for ϕJ(t)\phi_J(t) in (34). Applying this method using central differencing yields time-marching equations for each dynamical variable,

[Tu]{ϕ(n+1)}=(Δt)22({fu(n+1/2)}+{fu(n−1/2)})−(Δt)2[Su]−2[Tu]{ϕ(n)}−[Tu]{ϕ(n−1)}(36)[T_u]\{\phi^{(n+1)}\}=\frac{(\Delta t)^2}{2}\left(\{f_u^{(n+1/2)}\}+\{f_u^{(n-1/2)}\}\right)-(\Delta t)^2[S_u]-2[T_u]\{\phi^{(n)}\}-[T_u]\{\phi^{(n-1)}\} \tag*{(36)}

and

ϕJ(n+3/2)=2ϕJ(n+1/2)−Ic(Δt)2CJsin⁡(2eℏϕJ(n+1/2))−ϕJ(n−1/2)\phi_J^{(n+3/2)}=2\phi_J^{(n+1/2)}-\frac{I_c(\Delta t)^2}{C_J}\sin\left(\frac{2e}{\hbar}\phi_J^{(n+1/2)}\right)-\phi_J^{(n-1/2)}
−(Δt)24L1CJΔz(ϕM(n+1)+ϕM(n)−ϕM−1(n+1)−ϕM−1(n))-\frac{(\Delta t)^2}{4L_1C_J\Delta z}\left(\phi_M^{(n+1)}+\phi_M^{(n)}-\phi_{M-1}^{(n+1)}-\phi_{M-1}^{(n)}\right)
−(Δt)24L2CJΔz(ϕM+2(n+1)+ϕM+2(n)−ϕM+1(n+1)−ϕM+1(n)),(37)-\frac{(\Delta t)^2}{4L_2C_J\Delta z}\left(\phi_{M+2}^{(n+1)}+\phi_{M+2}^{(n)}-\phi_{M+1}^{(n+1)}-\phi_{M+1}^{(n)}\right), \tag*{(37)}

where ϕM(n)\phi_M^{(n)} is the flux at node MM and timestep nn. In formulating (36) and (37), averaging was used to approximate unknown values, i.e.: ϕM(n+1/2)≈12(ϕM(n+1)+ϕM(n))\phi_M^{(n+1/2)}\approx\frac{1}{2}\left(\phi_M^{(n+1)}+\phi_M^{(n)}\right). Notably, using central differencing to discretize (35) led to a time-marching equation for ϕJ(t)\phi_J(t) that can be solved linearly, despite the nonlinearity of (35). As a result, both dynamical variables can be evolved in time without the need for a nonlinear solver.

Full JTWPA Simulation

The method is extended from a single unit cell to a JTWPA by connecting NN unit cells with per-unit-length inductance LunL_{\mathrm{un}} and capacitance CunC_{\mathrm{un}} together at their endpoints, as shown in Figure 3. To drive the JTWPA, a voltage source Vs(t)V_s(t) with resistance RsR_s is connected to the input side through a transmission line with per-unit-length inductance LinL_{\mathrm{in}} and capacitance CinC_{\mathrm{in}}. On the output side, a load RlR_l is connected through another transmission line characterized by per-unit-length parameters LoutL_{\mathrm{out}} and CoutC_{\mathrm{out}} to terminate the simulation region. The source and load capacitors CsC_s and ClC_l are introduced to aid in filtering out low-frequency spurious modes that lead to late-time instability, which will be discussed in more detail in Section IV. The boundary conditions for the source and load are introduced in [32] as

Schematic of a JTWPA

Figure 3. Schematic of a JTWPA. Capacitive loads are introduced at the source and termination of the JTWPA to aid in filtering out low-frequency spurious modes that lead to late-time instability.

ϕ′(zs,t)=−LinRs[Vs(t)−ϕ˙(zs,t)],(38)\phi'(z_s,t)=-\frac{L_{\mathrm{in}}}{R_s}[V_s(t)-\dot{\phi}(z_s,t)], \tag*{(38)}
ϕ′(zl,t)=−LoutRlϕ˙(zl,t).(39)\phi'(z_l,t)=-\frac{L_{\mathrm{out}}}{R_l}\dot{\phi}(z_l,t). \tag*{(39)}

where zsz_s and zlz_l are the positions of the source and load. Since these terms contain a first derivative in time, the matrix equation of (31) needs to be modified to the form

[T]d2dt2{ϕ(t)}+[R]ddt{ϕ(t)}+[S]{ϕ(t)}={f(t)}.(40)[T]\frac{d^2}{dt^2}\{\phi(t)\}+[R]\frac{d}{dt}\{\phi(t)\}+[S]\{\phi(t)\}=\{f(t)\}. \tag*{(40)}

The matrices for the full JTWPA in this expression become

[T]=[Tin]+∑n[Tun]+[Tout],(41)[T]=[T_{\mathrm{in}}]+\sum_n[T_{\mathrm{un}}]+[T_{\mathrm{out}}], \tag*{(41)}
[R]ij=LinRsδi1δj1+LoutRlδiNeδjNe,(42)[R]_{ij}=\frac{L_{\mathrm{in}}}{R_s}\delta_{i1}\delta_{j1}+\frac{L_{\mathrm{out}}}{R_l}\delta_{iN_e}\delta_{jN_e}, \tag*{(42)}
[S]=[Sin]+∑n[Sun]+[Sout],(43)[S]=[S_{\mathrm{in}}]+\sum_n[S_{\mathrm{un}}]+[S_{\mathrm{out}}], \tag*{(43)}
{f(t)}i=LinRsVs(t)δi1+∑n{fun(t)},(44)\{f(t)\}_i=\frac{L_{\mathrm{in}}}{R_s}V_s(t)\delta_{i1}+\sum_n\{f_{\mathrm{un}}(t)\}, \tag*{(44)}

where node 1 occurs at the source, and node NeN_e is positioned at the load. Further, [Tun][T_{\mathrm{un}}], [Sun][S_{\mathrm{un}}], and {fun(t)}\{f_{\mathrm{un}}(t)\} refer to (32), (33), and (34) for unit cell nn. The input matrices have the form

[Tin]ij=LinCin∫zsz0Ni(z)Nj(z) dz[T_{\mathrm{in}}]_{ij}=L_{\mathrm{in}}C_{\mathrm{in}}\int_{z_s}^{z_0}N_i(z)N_j(z)\,dz
[Tin]ij=+LinCs(δ11−δ12−δ21+δ22),(45)\phantom{[T_{\mathrm{in}}]_{ij}=}+L_{\mathrm{in}}C_s(\delta_{11}-\delta_{12}-\delta_{21}+\delta_{22}), \tag*{(45)}
[Sin]ij=∫zsz0Ni′(z)Nj′(z) dz,(46)[S_{\mathrm{in}}]_{ij}=\int_{z_s}^{z_0}N_i'(z)N_j'(z)\,dz, \tag*{(46)}

where z0z_0 is the starting position of the first unit cell. The output matrices can be expressed as

[Tout]ij=LoutCout∫zezlNi(z)Nj(z) dz[T_{\mathrm{out}}]_{ij}=L_{\mathrm{out}}C_{\mathrm{out}}\int_{z_e}^{z_l}N_i(z)N_j(z)\,dz
[Tout]ij=+LoutCl(δ(Ne−1)(Ne−1)−δ(Ne−1)Ne−δNe(Ne−1)+δNeNe),(47)\phantom{[T_{\mathrm{out}}]_{ij}=}+L_{\mathrm{out}}C_l\left(\delta_{(N_e-1)(N_e-1)}-\delta_{(N_e-1)N_e}-\delta_{N_e(N_e-1)}+\delta_{N_eN_e}\right), \tag*{(47)}
[Sout]ij=∫zezlNi′(z)Nj′(z) dz.(48)[S_{\mathrm{out}}]_{ij}=\int_{z_e}^{z_l}N_i'(z)N_j'(z)\,dz. \tag*{(48)}

where zez_{\mathrm{e}} is the end position of the final unit cell. With the matrices in (40) constructed, central differencing can be applied to produce a time-marching equation,

([T]+Δt2[R]){ϕ(n+1)}=(Δt)2{f(n)}+(Δt)22({f(n+1/2)}+{f(n−1/2)})−((Δt)2[S]−2[T]){ϕ(n)}−([T]−Δt2[R]){ϕ(n−1)}.(49)\left([T]+\frac{\Delta t}{2}[R]\right)\{\phi^{(n+1)}\}=(\Delta t)^2\{f^{(n)}\}+\frac{(\Delta t)^2}{2}\left(\{f^{(n+1/2)}\}+\{f^{(n-1/2)}\}\right)-\left((\Delta t)^2[S]-2[T]\right)\{\phi^{(n)}\}-\left([T]-\frac{\Delta t}{2}[R]\right)\{\phi^{(n-1)}\}. \tag*{(49)}

The time marching equation for ϕJ(t)\phi_J(t) in (37) is unchanged for the full JTWPA, but now must be independently solved for each Josephson junction in the circuit.

RPM JTWPA

The inclusion of resonant loads introduces the resonator flux ϕr(t)\phi_r(t) as a third dynamical variable for an RPM JTWPA. However, since (29) is linear, there is no need to incorporate leap-frogging to evolve ϕr(t)\phi_r(t) in time. Instead, the resonator flux ϕr(t)\phi_r(t) can be solved simultaneously with ϕ(z,t)\phi(z,t). To accomodate this change, we modify (40) to the form

\begin{bmatrix} T_{\ell\ell} & T_{\ell r} \\ T_{r\ell} & T_{rr} \end{bmatrix}\frac{d^2}{dt^2}\begin{Bmatrix}\phi(t)\\\phi_r(t)\end{Bmatrix}+\begin{bmatrix} R_{\ell\ell} & 0 \\ 0 & 0 \end{bmatrix}\frac{d}{dt}\begin{Bmatrix}\phi(t)\\\phi_r(t)\end{Bmatrix}+\begin{bmatrix} S_{\ell\ell} & 0 \\ 0 & S_{rr} \end{bmatrix}\begin{Bmatrix}\phi(t)\\\phi_r(t)\end{Bmatrix}=\begin{Bmatrix}f(t)\\0\end{Bmatrix}, \tag*{(50)}\end{bmatrix}

where {ϕr(t)}i=ϕr,i(t)\{\phi_r(t)\}_i=\phi_{r,i}(t), or the flux in the resonant load of the iith unit cell. Further, [R]ℓℓ[R]_{\ell\ell}, [S]ℓℓ[S]_{\ell\ell}, and {f(t)}\{f(t)\} are equivalent to their counterparts in (42) to (44) for the full JTWPA without RPM. From (23) to (29), the remaining matrices can be expressed as

[Tℓℓ]ij=[T]ij+∑nLunCcnδiRnδjRn(51)[T_{\ell\ell}]_{ij}=[T]_{ij}+\sum_n L_{un}C_{cn}\delta_{iR_n}\delta_{jR_n} \tag*{(51)}
[Tℓr]ij=∑n−LunCcnδiRnδjn,(52)[T_{\ell r}]_{ij}=\sum_n-L_{un}C_{cn}\delta_{iR_n}\delta_{jn}, \tag*{(52)}
[Trℓ]ij=∑n−LrnCcnδinδjRn,(53)[T_{r\ell}]_{ij}=\sum_n-L_{rn}C_{cn}\delta_{in}\delta_{jR_n}, \tag*{(53)}
[Trr]ij=∑nLrn(Ccn+Crn)δinδjn,(54)[T_{rr}]_{ij}=\sum_n L_{rn}(C_{cn}+C_{rn})\delta_{in}\delta_{jn}, \tag*{(54)}
[Srr]=∑nδinδjn,(55)[S_{rr}]=\sum_n\delta_{in}\delta_{jn}, \tag*{(55)}

where each node RnR_n is where the resonant load is located in the nnth unit cell, and LrnL_{rn}, CcnC_{cn}, and CrnC_{rn} are LrL_r, CcC_c, and CrC_r for that unit cell.

The time-marching expression for the RPM JTWPA is the same as the equation for the basic JTWPA in (49), but with the matrices and vectors modified to the block form of (50). The time-marching equation for each ϕJ(t)\phi_J(t) in (37) is unchanged by the addition of resonant loads.

Results

Here, we present results to validate the numerical method and exhibit its utility. In Section IV-A, the gain of a basic JTWPA design is calculated from the results of the numerical method and compared to analytical results. This procedure is extended to a JTWPA architecture using RPM in Section IV-B, which also includes a discussion of the strategy used to inhibit late-time instability. In Section IV-C, the method is applied to simulate the effect of variations in Josephson junction area across a JTWPA. Further, the resonant loads are varied in Section IV-D to demonstrate another potential application.

Basic JTWPA Gain Curve

The method is initially validated by comparing the analytical results from [25] for the basic JTWPA design presented in [12]. The specifications for this JTWPA are detailed in Table I. Notably, the inductance of each unit cell in this design is assumed to be dominated by LJL_J, meaning Lu≈0L_u\approx0. This assumption is not suitable for our numerical method, so a small per-unit-length inductance of Lu=1 μH/mL_u=1\ \mu\mathrm{H}/\mathrm{m} is added to the transmission lines. This small LuL_u was found to have a minimal impact on the numerical results, while also not modifying the FETD stability constraint to a problematic extent. To match the input and output transmission lines to the JTWPA, the contributions of the Josephson junctions need to be accounted for, so Lin=Lout=Lu+LJ/aL_{\mathrm{in}}=L_{\mathrm{out}}=L_u+L_J/a.

ParameterValue
Junction critical current (IcI_c)3.29 μA\mu\mathrm{A}
Josephson capacitance (CJC_J)329 fF
Per-unit-length capacitance (CuC_u)3.9 nF/m
Unit cell length (aa)10 μ\mum
Number of unit cells (NN)2000
Pump current magnitude (IpI_p)1.645 μA\mu\mathrm{A}
Pump frequency (fpf_p)5.970 GHz

Table 1. Parameters for the basic JTWPA architecture from [12]

The JTWPA is driven with a voltage source

Vs(t)=2IpZ0W(t)(sin⁡(ωpt)+αsin⁡(ωst)),(56)V_{\mathrm{s}}(t)=2I_{\mathrm{p}}Z_{0}W(t)\left(\sin(\omega_{\mathrm{p}}t)+\alpha\sin(\omega_{\mathrm{s}}t)\right), \tag*{(56)}

where ωs\omega_{\mathrm{s}} is the signal frequency, ωp\omega_{\mathrm{p}} is the pump frequency, and Z0=(aLu+LJ)/aCuZ_{0}=(aL_{\mathrm{u}}+L_{\mathrm{J}})/aC_{\mathrm{u}} is the intrinsic impedance of the transmission line. The ratio between the signal and pump amplitudes is determined by α\alpha, which is set to α=1/20\alpha=1/20 to prevent pump depletion from compressing gain. Further, W(t)W(t) is an exponential tapering function with a total width of 50 ns. While analytical models typically assume the pump and signal are monochromatic, due to our method operating in the time domain a finite bandwidth is necessary to keep the total simulation duration at a reasonable level.

The gain GG is calculated from simulation results via

G=10log⁡(∫fLfH∣ϕo(f)∣2 df∫fLfH∣ϕi(f)∣2 df),(57)G=10\log\left(\frac{\int_{f_{\mathrm{L}}}^{f_{\mathrm{H}}}\lvert\phi_{\mathrm{o}}(f)\rvert^{2}\,\mathrm{d}f}{\int_{f_{\mathrm{L}}}^{f_{\mathrm{H}}}\lvert\phi_{\mathrm{i}}(f)\rvert^{2}\,\mathrm{d}f}\right), \tag*{(57)}

where ϕi(f)\phi_{\mathrm{i}}(f) and ϕo(f)\phi_{\mathrm{o}}(f) are the Fourier transform of the node flux at the input and output respectively. The integration bandwidth is determined by fLf_{\mathrm{L}} and fHf_{\mathrm{H}}, which are set to fL=fs−0.5f_{\mathrm{L}}=f_{\mathrm{s}}-0.5 GHz and fH=fs+0.5f_{\mathrm{H}}=f_{\mathrm{s}}+0.5 GHz. Wideband integration is necessary due to self-mixing of the signal, which spreads energy from its center frequency over a wider frequency range. In an experimental setting, the signal is typically several orders of magnitude smaller than the critical current, which limits this self-mixing process. In the numerical method, this process becomes relevant due to the relatively large signal-to-pump ratio, which was chosen to allow an increased pump bandwidth that helps reduce the total simulation duration to keep the method more efficient.

Using (57), the gain is extracted from numerical simulations for the architecture of Table I and compared to the analytical result of [25] in Figure 4. At signal frequencies between 5.3 GHz and 6.6 GHz, the integration bandwidth overlaps with the pump and induces error in the gain calculation. To allow the gain to be extracted in this region, a quadratic fit is used to interpolate it from surrounding points. While interpolation is necessary for this topology, alternative JTWPA designs using three-wave mixing remove the pump from the operating bandwidth, allowing this process to be avoided. These architectures will be discussed in more detail in § 5 and considered in future work.

Gain of the architecture of Table I

Figure 4. Gain of the architecture of Table I calculated using the analytical method of [25] and the numerical method of this work. Points between 5.3 GHz and 6.6 GHz are interpolated using a quadratic fit.

Image not extracted

Figure 5. Gain of the architecture of Table II calculated using the analytical method of [25] and the numerical method of this work. Points between 5.3 GHz and 6.6 GHz are interpolated using a fourth-order polynomial fit.

RPM JTWPA Gain Curve

The method is also validated using the RPM JTWPA design in [12], for which the relevant parameter values are listed in Table II. For this architecture, some additional changes to the voltage source described in (56) must be made. First of all, the addition of the resonant loads shifts the average intrinsic impedance of the JTWPA to Z0=(aLu+LJ)/(aCu+Cc)Z_{0}=(aL_{\mathrm{u}}+L_{\mathrm{J}})/(aC_{\mathrm{u}}+C_{\mathrm{c}}). To compensate for this, the per-unit-length capacitance of the input and output transmission lines must also be adjusted to Cin=Cout=Cu+Cc/aC_{\mathrm{in}}=C_{\mathrm{out}}=C_{\mathrm{u}}+C_{\mathrm{c}}/a. Additionally, the increased gain of this design requires that the signal-to-pump ratio is reduced to α=1/50\alpha=1/50 to minimize the effect of pump depletion. Finally, the pump bandwidth must be reduced for this architecture due to the addition of the resonant loads. To correct for mismatch between the signal and pump phases, the resonant loads introduce a pole at 6 GHz, where its proximity to the pump leads to a phase shift. However, if the pump is too wideband, this proximity means the sidebands of the pump will overlap with the pole, resulting in deformation. In an experimental setting, the pump is typically narrowband enough that this is not an issue. However, the shortened envelope used in the numerical method increases the pump bandwidth, causing it to extend into the pole. To minimize this issue, the width of the tapering function W(t)W(t) was increased to 250 ns.

While the narrowband pump is necessary in this architecture, increasing the duration of W(t)W(t) requires an equivalent increase in total simulation time. As a result, low-frequency spurious solutions are given time to grow and eventually produce instability. To address this issue, we applied the correction method detailed in [35]. This method periodically isolates and removes the electrostatic portion of the solution to prevent buildup of spurious solutions. To separate this

ParameterValue
Junction critical current (IcI_{\mathrm{c}})3.29 μ\muA
Josephson capacitance (CJC_{\mathrm{J}})329 fF
Per-unit-length capacitance (CuC_{\mathrm{u}})3.9 nF/m
Unit cell length (aa)10 μ\mum
Resonant load coupling capacitance (CcC_{\mathrm{c}})10 fF
Resonant load capacitance (CrC_{\mathrm{r}})7.036 pF
Resonant load inductance (LrL_{\mathrm{r}})100 pH
Number of unit cells (NN)2000
Pump current magnitude (IpI_{\mathrm{p}})1.645 μ\muA
Pump frequency (fpf_{\mathrm{p}})5.970 GHz

Table 2. Parameters for the RPM JTWPA architecture from [12]

electrostatic component, a basis for the nullspace of [S][S] must be known. For the 1D system, each vector in the basis is constant over a single segment of transmission line terminated by Josephson junctions or capacitors, and zero elsewhere.

Importantly, the correction method of [35] is designed to be performed on lossless systems. While the JTWPA itself is lossless, RsR_s and RlR_l introduce loss which can produce inaccuracy in the correction process. To avoid needing to apply the method in these regions, capacitors CsC_s and ClC_l are introduced. These capacitors separate the source and load from the input and output transmission lines, which removes them from the corresponding nullspace vectors. As a result, the correction method can be applied to the input and output transmission lines without also applying it to the source and load. To prevent instability from building up in the uncorrected regions and spreading to the JTWPA, the capacitors are sized at Cs=Cl=100 pFC_s = C_l = 100\ \mathrm{pF}. At this value, they present a high impedance to spurious solutions to inhibit their growth. However, the impedance at the operating frequency range is small enough that the capacitors do not have a significant impact on the accuracy of the method.

The gain extrapolated from numerical results for the Table II JTWPA using (57) is compared against the analytical result from [25] in Figure 4. For this architecture, a fourth-order polynomial interpolating function was found to provide good agreement with the analytical results in the vicinity of the pump. At 6 GHz, the pole introduced by the resonant loads produces a large decrease in gain. Due to the use of interpolation in this region, this effect is not modeled by the numerical method.

Josephson Junction Area Variations

To demonstrate a potential application of the numerical method, we investigated the effect of variations in Josephson junction area on JTWPA gain. The junction area AA is related to the Josephson inductance and capacitance by LJ∝AL_J \propto A and CJ∝1/AC_J \propto1/A [36]. Figure 6(a) demonstrates the result of independently varying the area of each Josephson junction in the Table I JTWPA on a normal distribution with mean AA and standard deviation σA\sigma_A. For σA=0.03A\sigma_A = 0.03A, the impact is relatively minor, and gain remains close to the analytical result. The differences are more pronounced for σA=0.06A\sigma_A = 0.06A and σA=0.09A\sigma_A = 0.09A, especially at signal frequencies greater than the pump frequency, where substantial gain ripples occur.

Gain curves for variations in Josephson junction area

Figure 6. Gain of the Table I JTWPA with variations in Josephson junction area. Each curve is shaded between the minimum and maximum outcomes of 50 simulations. The area of each junction in the JTWPA is independently selected on a normal distribution with standard deviation σA\sigma_A. The mean of the distribution is (a) constant, or linearly varied by (b) 0.05A0.05A, (c) 0.10A0.10A, or (d) 0.15A0.15A across the JTWPA.

In addition to entirely random variations, significant gradients in Josephson junction area across a JTWPA have been observed during device manufacturing [37]. To model these gradients, we varied the mean of the normal distribution as a linear function of position along the JTWPA. In Figure 6(b), the gradient is set so that the mean area of the final Josephson junction in the JTWPA is 0.05A0.05A higher than the first junction. In Figure 6(c) and Figure 6(d), this procedure was repeated with total gradients of 0.10A0.10A and 0.15A0.15A respectively. These results indicate that increasing the gradient reduces gain across the bandwidth of the device.

The impact of junction area fluctuations was also tested for the JTWPA with RPM in Figure 7, where a gradient of 0.05A0.05A was used. In this case, the presence of the resonant loads leads to additional gain ripples near the pump frequency.

Gain of the Table II JTWPA with variations in Josephson junction area

Figure 7. Gain of the Table II JTWPA with variations in Josephson junction area. Each curve is shaded between the minimum and maximum outcomes of 30 simulations. The area of the Josephson junctions are selected from a normal distribution with standard deviation σA\sigma_A, and with its mean linearly varied by 0.05A0.05A across the JTWPA.

Resonant Load Variations

In addition to the Josephson junction area, the method was applied to investigate how small variations in the values of the resonant load components can affect JTWPA gain. While these components do not suffer from the same manufacturing inconsistencies as Josephson junctions, the delicate nature of the phase matching condition means small changes can severely impact performance. To demonstrate this, we varied LrL_r and CrC_r for each load on normal distributions with standard deviations σLr\sigma_{L_r} and σCr\sigma_{C_r}. The mean value for each distribution matches the component value used in Table II. The result of applying these variations are displayed in Figure 8 for several values of σLr\sigma_{L_r} and σCr\sigma_{C_r}. For σLr(Cr)=0.0010Lr(Cr)\sigma_{L_r(C_r)} = 0.0010L_r(C_r) and σLr(Cr)=0.0015Lr(Cr)\sigma_{L_r(C_r)} = 0.0015L_r(C_r), the effect of these variations is mostly negligible, indicating that changes in the pump phase due to the shifting resonance roughly cancel across the JTWPA. However, the impact at σLr(Cr)=0.0020Lr(Cr)\sigma_{L_r(C_r)} = 0.0020L_r(C_r), is clearly much more significant. At this value, the standard deviation is large enough that the pole may be shifted so that it overlaps with the pump bandwidth. As a result, the pump is partially absorbed by the resonant load, leading to the gain drop-off observed in the plot.

Gain of the Table II JTWPA with variations in the resonant loads

Figure 8. Gain of the Table II JTWPA with variations in the resonant loads. Each curve is shaded between the minimum and maximum outcomes of 30 simulations. For each load, LrL_r and CrC_r are independently chosen from normal distributions with standard deviations σLr\sigma_{L_r} and σCr\sigma_{C_r}, and mean values matching those in Table II.

Conclusion

In this work, a numerical method for modeling JT PWAs was developed to overcome the limitations of analytical models. Rather than using a continuum approximation as typically done in analytical approaches, the Josephson junctions and transmission lines were treated as coupled subsystems. As a result, components can easily be individually varied to examine the impact of manufacturing tolerances on device performance. To linearize the time-marching equations for the dynamical variables, we employed a leap-frog time-marching procedure. Gain was extracted from simulations of JTWPA designs with and without RPM and compared to analytical results to validate the method. Further, the impact of variations in Josephson junctions and resonant load components was investigated using the method.

Future work will expand the method to support additional JTWPA architectures and explore other effects not captured by analytical models. Modern JTWPA designs integrate Josephson junction-based circuit elements such as superconducting quantum interference devices (SQUIDs) [17] and superconducting nonlinear asymmetric inductive elements (SNAILs) [18] to leverage their additional degrees of freedom. These components can also be used to develop three-wave mixing JTWPas which are inherently phase matched [24], and can use a flux-driven topology to achieve high dynamic range [37], [38]. However, the performance of these JTWPas is heavily impacted by the generation of harmonics, an effect which is not captured by analytical models [27]. By incorporating components such as SQUIDs and SNAILs into our multiphysics method, this issue can be explored in greater detail.

In addition to supporting more JTWPA architectures, future work will extend the numerical method to a 3D full-wave description. While analytical device models are inherently limited to 1D, the need to extract circuit parameters of JTWPA geometries limits the accuracy of those methods and adds a significant pre-processing step. Meanwhile, a 3D numerical method would be capable of modeling the intended and non-ideal behavior of JTWPAs with unprecedented accuracy.

References

Biographies

Samuel T. Elkin

Samuel T. Elkin (S’22) received the B.S degree in electrical engineering from Purdue University, West Lafayette, IN, USA, in 2022. From 2022 to 2023, he worked at Indesign, LLC., Indianapolis, IN, USA, as an electrical engineer. He is currently pursuing a Ph.D. degree in electrical and computer engineering at Purdue University. His current research interests are in multiphysics modeling of superconducting circuit quantum devices.

Michael Haider

Michael Haider (Member, IEEE) received the B.Sc., M.Sc., and Ph.D. (summa cum laude) degrees in electrical and computer engineering from the Technical University of Munich (TUM), Garching, Germany, in 2014, 2016, and 2019, respectively. From 2016 to 2019, he was a Research Associate with the Institute of Nanoelectronics, TUM, where his research focused on quantum circuit theory, wireless power transfer, and stochastic electromagnetic fields. In 2019, he joined the Computational Photonics Group, TUM, as a Post-Doctoral Researcher, where he became a permanent staff Scientist in October 2023. He has coauthored over 100 scientific papers in refereed journals and conference proceedings. His current research interests include the modeling of charge carrier transport in quantum cascade lasers, the dynamical modeling of quantum devices within a generalized Maxwell-Bloch framework, and quantum circuit theoretical models applied to Josephson traveling-wave parametric amplifiers. Dr. Haider received the 2020 VDE ITG Dissertationspreis for his outstanding Ph.D. thesis on Investigations of Stochastic Electromagnetic Fields.

Thomas E. Roth

Thomas E. Roth (Member, IEEE) received the B.S. degrees in electrical engineering and computer engineering from the Missouri University of Science and Technology, Rolla, MO, USA, in 2015, and the M.S. and Ph.D. degrees in electrical and computer engineering from the University of Illinois at Urbana-Champaign, Champaign, IL, USA, in 2017 and 2020, respectively. He is currently an Assistant Professor with the Elmore Family School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA. Prior to joining Purdue, he was a Senior Member of the Technical Staff with Sandia National Laboratories, Albuquerque, NM, USA. He has received five best paper or young scientist awards from conferences, and the Ruth and Joel Spira Outstanding Teacher Award from Purdue University. His current research interests include multiscale and multiphysics computational electromagnetics techniques, particularly for rigorously analyzing quantum information processing devices.

References

  1. [1]F. Arute, K. Arya, R. Babbush, D. Bacon, J. Bardin, R. Barends, R. Biswas, S. Boixo, F. Brandao, D. Buell et al., “Quantum supremacy using a programmable superconducting processor,” Nature, vol. 574, p. 505–510, Oct 2019.
  2. [2]Y. Wu, W.-S. Bao, S. Cao, F. Chen, M.-C. Chen, X. Chen, T.-H. Chung, H. Deng, Y. Du, D. Fan et al., “Strong quantum computational advantage using a superconducting quantum processor,” Physical Review Letters, vol. 127, no. 18, p. 180501, 2021.
  3. [3]P. Jurcevic, A. Javadi-Abhari, L. S. Bishop, I. Lauer, D. F. Bogorin, M. Brink, L. Capelluto, O. Günlük, T. Itoko, N. Kanazawa et al., “Demonstration of quantum volume 64 on a superconducting quantum computing system,” Quantum Science and Technology, vol. 6, no. 2, p. 025020, 2021.
  4. [4]T. E. Roth, R. Ma, and W. C. Chew, “The transmon qubit for electromagnetics engineers: An introduction,” IEEE Antennas and Propagation Magazine, vol. 65, no. 2, pp. 8–20, 2023.
  5. [5]P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gustavsson, and W. D. Oliver, “A quantum engineer’s guide to superconducting qubits,” Applied physics reviews, vol. 6, no. 2, 2019.DOI
  6. [6]T. Walter, P. Kurpiers, S. Gasparinetti, P. Magnard, A. Potočnik, Y. Salathe, M. Pechal, M. Mondal, M. Oppliger, C. Eichler, and A. Wallraff, “Rapid high-fidelity single-shot dispersive readout of superconducting qubits,” Phys. Rev. Appl., vol. 7, p. 054020, May 2017.
  7. [7]J. Aumentado, “Superconducting parametric amplifiers: The state of the art in Josephson parametric amplifiers,” IEEE Microwave Magazine, vol. 21, no. 8, pp. 45–59, Jul 2020.
  8. [8]H. Zimmer, “Parametric amplification of microwaves in superconducting Josephson tunnel junctions,” Applied Physics Letters, vol. 10, no. 7, pp. 193–195, 04 1967.
  9. [9]P. Russer, “Parametric amplification with Josephson junctions,” AEÜ Archiv der Elektrotechnischen Übertragung, vol. 23, no. 8, pp. 417–420, 1969.
  10. [10]M. A. Castellanos-Beltran and K. W. Lehnert, “Widely tunable parametric amplifier based on a superconducting quantum interference device array resonator,” Applied Physics Letters, vol. 91, no. 8, p. 083509, Aug 2007.
  11. [11]M. Esposito, A. Ranadive, L. Planat, and N. Roch, “Perspective on traveling wave microwave parametric amplifiers,” Applied Physics Letters, vol. 119, no. 12, p. 120501, Sep 2021.
  12. [12]K. O’Brien, C. Macklin, I. Siddiqi, and X. Zhang, “Resonant phase matching of Josephson junction traveling wave parametric amplifiers,” Phys. Rev. Lett., vol. 113, p. 157001, Oct 2014.
  13. [13]C. Macklin, K. O’Brien, D. Hover, M. E. Schwartz, V. Bolkhovsky, X. Zhang, W. D. Oliver, and I. Siddiqi, “A near-quantum-limited Josephson traveling-wave parametric amplifier,” Science, vol. 350, no. 6258, pp. 307–310, Sep 2015.
  14. [14]C. White, J. Y. Mutus, I.-C. Hoi, R. Barends, B. Campbell, Y. Chen, Z. Chen, B. Chiaro, A. Dunsworth, E. Jeffrey et al., “Traveling wave parametric amplifier with Josephson junctions using minimal resonator phase matching,” Applied Physics Letters, vol. 106, no. 24, p. 242601, Jun 2015.
  15. [15]L. A. Grimsmo and A. Blais, “Squeezing and quantum state engineering with Josephson travelling wave amplifiers,” npj Quantum Information, vol. 3, p. 20, Jun 2017.
  16. [16]J. Y. Qiu, A. Grimsmo, K. Peng, B. Kannan, B. Lehnard, Y. Sung, P. Krantz, V. Bolkhovsky, G. Calusine, D. Kim et al., “Broadband squeezed microwaves and amplification with a Josephson travelling-wave parametric amplifier,” Nature Physics, vol. 19, pp. 706–713, May 2023.
  17. [17]L. Planat, A. Ranadive, R. Dassonneville, J. Puertas Martínez, S. Léger, C. Naud, O. Buisson, W. Hasch-Guichard, D. M. Baskо, and N. Roch, “Photonic-crystal Josephson traveling-wave parametric amplifier,” Phys. Rev. X, vol. 10, p. 021021, Apr 2020.DOI
  18. [18]A. Ranadive, M. Esposito, L. Planat, E. Bonet, C. Naud, O. Buisson, W. Guichard, and N. Roch, “Kerr reversal in Josephson meta-material and traveling wave parametric amplification,” Scientific Reports, vol. 13, Apr 2022.
  19. [19]A. Osman, J. Simon, A. Bengtsson, S. Kosen, P. Krantz, D. P. Lozano, M. Scigliuzzo, P. Delsing, J. Bylander, and A. Fadavi Roudsari, “Simplified Josephson-junction fabrication process for reproducibly high-performance superconducting qubits,” Applied Physics Letters, vol. 118, no. 6, p. 064002, Feb 2021.
  20. [20]A. Pishchimova, N. Smirnov, D. Ezenkova, E. Krivko, E. Zikiy, D. Moskalev, A. Ivanov, N. Korshakov, and I. Rodionov, “Improving Josephson junction reproducibility for superconducting quantum circuits: Junction area fluctuation,” Scientific Reports, vol. 13, Apr 2023.
  21. [21]J. Bylander, “Reproducible Josephson junctions are important for scalable quantum technologies,” Superconductor Science and Technology, vol. 33, no. 11, p. 110501, Sep 2020.
  22. [22]D. C. Feng, M. Vahidpour, Y. Mohan, N. Sharac, T. Whyland, S. Stanwyck, G. Ramachandran, and M. Selvanayagam, “Design and measurement of a Josephson traveling wave parametric amplifier fabricated in a superconducting qubit process,” in 2020 IEEE/MIT-S International Microwave Symposium (IMS), 2020, pp. 940–943.
  23. [23]O. Yaakobi, L. Friedland, C. Macklin, and I. Siddiqi, “Parametric amplification in Josephson junction embedded transmission lines,” Phys. Rev. B, vol. 87, p. 144301, Apr 2013.
  24. [24]A. B. Zorin, “Josephson traveling-wave parametric amplifier with three-wave mixing,” Phys. Rev. Appl., vol. 6, p. 034006, Sep 2016.
  25. [25]Y. Yuan, M. Haider, J. A. Russer, P. Russer, and C. Jirausek, “Circuit quantum electrodynamic model of dissipative-dispersive Josephson traveling-wave parametric amplifiers,” Phys. Rev. A, vol. 107, p. 022612, Feb 2023.
  26. [26]M. T. Bell and A. Samolov, “Traveling-wave parametric amplifier based on a chain of coupled asymmetric squids,” Phys. Rev. Appl., vol. 4, p. 024014, Aug 2015.
  27. [27]T. Dixon, J. Dunstan, G. Long, J. Williams, P. Meeson, and C. Shelly, “Capturing complex behavior in Josephson traveling-wave parametric amplifiers,” Phys. Rev. Appl., vol. 14, p. 034058, Sep 2020.
  28. [28]S. O. Peataín, T. Dixon, P. J. Meeson, J. M. Williams, S. Kafanov, and Y. A. Pashkin, “Simulating the effects of fabrication tolerance on the performance of Josephson traveling wave parametric amplifiers,” Superconductor Science and Technology, vol. 36, no. 4, p. 045017, Mar 2023.
  29. [29]C. Guarcello, G. Avallone, C. Barone, M. Borghesi, S. Capelli, G. Carapella, A. P. Caricato, I. Carusotto, A. Cian, D. Di Gioacchino et al., “Modeling of Josephson traveling wave parametric amplifiers,” IEEE Transactions on Applied Superconductivity, vol. 33, no. 1, pp. 1–7, Jan 2023.
  30. [30]S. T. Elkin, M. Haider, and T. E. Roth, “Hybrid numerical modeling method for Josephson traveling wave parametric amplifiers,” in 2024 IEEE International Symposium on Antennas and Propagation and USNC-URSI Radio Science Meeting (AP-S/URSI). (accepted for publication), 2024.
  31. [31]M. Stone and P. Goldbart, Mathematics for Physics: A Guided Tour for Graduate Students. Cambridge University Press, 2009.
  32. [32]T. E. Roth and S. T. Elkin, “Maxwell-Schrödinger modeling of a superconducting qubit coupled to a transmission line network,” IEEE Journal on Multiscale and Multiphysics Computational Techniques, vol. 9, pp. 61–74, Jan 2024.
  33. [33]S. M. Girvin, “Circuit QED: Superconducting qubits coupled to microwave photons,” in Quantum Machines: Measurement and Control of Engineered Quantum Systems. Oxford University Press, 2014, pp. 113–256.
  34. [34]J.-M. Jin, Theory and Computation of Electromagnetic Fields. John Wiley & Sons, Ltd., 2010.DOI
  35. [35]R. A. Chilton and R. Lee, “The discrete origin of FETD-Newmark late time instability, and a correction scheme,” Journal of Computational Physics, vol. 224, no. 2, pp. 1293–1306, Jun 2007.
  36. [36]M. Tinkham, Introduction to Superconductivity: Second Edition, ser. Dover Books on Physics. Dover Publications, 2004.
  37. [37]A. Zorin, “Flux-driven Josephson traveling-wave parametric amplifier,” Phys. Rev. Appl., vol. 12, p. 044051, Oct 2019.DOI
  38. [38]M. Haider, “A black-box quantum model for superconducting traveling-wave parametric amplifiers,” IEEE Transactions on Microwave Theory and Techniques, pp. 1–15, Jan 2024.DOI

Paper details

vlnc:W_saavdu3p
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents