iToffoli Gate: Phase-Modified Three-Qubit Operation

• iToffoli gate is a phase-twisted three-qubit operation that replaces the standard X flip with an iX or -iX target flip, introducing a conditional phase shift.
• Native implementations in superconducting, silicon spin, and trapped-ion systems reduce circuit depth and improve fidelity through direct three-qubit interactions.
• Its unique phase characteristics enable efficient conversion to standard Toffoli or CCZ operations, lowering compilation overhead in quantum algorithms.

An iToffoli gate is a three-qubit Toffoli-family operation in which the activated target flip carries a nontrivial phase, so the nontrivial two-dimensional block is $iX$, $-iX$, or an equivalent phase-dressed form rather than the real $X$ of the standard controlled-controlled-NOT. In the recent literature, the term is used for several closely related conventions: a controlled-controlled-$(iX)$ gate on the $|11\rangle$ control sector, a controlled-controlled-$({-}iX)$ gate, and an open-control implementation activated on $|00\rangle$ instead of $|11\rangle$ (Tounsi et al., 2023, Rasmussen et al., 2019, Kim et al., 2021, Baker et al., 2021). The gate is important because native three-qubit implementations can reduce depth, entangling-gate count, and compilation overhead relative to decompositions into one- and two-qubit gates, and because the extra phase can be either a resource or a correction burden depending on the algorithm and hardware (Sun et al., 2023, Tounsi et al., 2023).

1. Definition and phase conventions

The standard Toffoli gate is the controlled-controlled-NOT $CCX$: it flips the target iff both controls are $1$. A common iToffoli convention replaces the activated $-iX$0 by $-iX$1. In the computational basis

$-iX$2

one explicit form is

$-iX$3

so $-iX$4, $-iX$5, and all other basis states are unchanged (Tounsi et al., 2023).

Other papers use a $-iX$6 convention rather than $-iX$7. In the strong-Ising single-step proposal, the authors define the i-Toffoli as the operation that acts as $-iX$8 on the target iff all controls are in $-iX$9 (Rasmussen et al., 2019). In the dispersively coupled superconducting proposal, the active subspace is $X$0, and the gate swaps those states with a factor $X$1 (Baker et al., 2021). In the fixed-frequency cross-resonance realization, the activated branch is chosen by open controls, so the target undergoes $X$2 when the two controls are in $X$3 rather than $X$4 (Kim et al., 2021).

A further convention issue is group structure. In the Fibonacci anyon compilation work, the Toffoli target operation is replaced by $X$5 because the braid representations of three Fibonacci anyons naturally live in an $X$6 structure, and the paper states that “The complex phase $X$7 is essential for maintaining gates in the $X$8 group” (Tounsi et al., 2023). This makes the iToffoli not merely a notational variant, but a form naturally adapted to the underlying representation theory.

2. Algebraic structure and relation to standard Toffoli

The crucial algebraic point is that an iToffoli is not generally equivalent to standard Toffoli by a harmless global phase. In the trapped-ion treatment, the gate is written as

$X$9

and the paper emphasizes that in the controlled setting the factor $(iX)$0 appears only in the selected control sector, so it becomes a relative phase between sectors rather than a global phase (Goel et al., 2021). Equivalently,

$(iX)$1

which exhibits iToffoli as standard Toffoli multiplied by a multi-controlled phase (Goel et al., 2021).

Several works exploit this structure to convert between iToffoli and standard Toffoli. In the fixed-frequency superconducting implementation, the iToffoli differs from the standard Toffoli only by a controlled-phase gate between the two controls, “which imparts a $(iX)$2 phase shift to $(iX)$3” (Kim et al., 2021). In the silicon-spin protocol, the native gate is an $(iX)$4-Toffoli and the full standard Toffoli is obtained by combining it with a compensating $(iX)$5 C-Phase and, if needed, SWAPs when a different qubit is to be the target (Gullans et al., 2019). In the strong-Ising proposal, a conventional Toffoli can be obtained from the native i-Toffoli using two applications of the primitive plus an ancilla and Hadamards (Rasmussen et al., 2019).

A broader mathematical family appears in the tetrahedron-equation construction. The paper introduces

$(iX)$6

and states that this coincides with the usual Toffoli at $(iX)$7 (Sinha et al., 2024). A plausible identification is that $(iX)$8 yields an iToffoli-like controlled-controlled-$(iX)$9 gate, although the paper itself presents this as a phase-twisted Toffoli family rather than by the specific name “iToffoli” (Sinha et al., 2024).

3. Native and single-step implementations in solid-state systems

A prominent experimental realization is the fixed-frequency superconducting-qubit iToffoli based on simultaneous cross-resonance driving in a linear chain $|11\rangle$0. The effective Hamiltonian is

$|11\rangle$1

with calibration condition

$|11\rangle$2

The gate is implemented by simultaneous microwave pulses at the target frequency, giving a $|11\rangle$3 rotation of the target for the selected $|11\rangle$4 control state and $|11\rangle$5 rotations otherwise. The reported gate duration is $|11\rangle$6 ns, and cycle benchmarking gives a SPAM-free process fidelity of $|11\rangle$7 (Kim et al., 2021).

A different superconducting proposal uses three transmons and two tunable couplers in the dispersive regime. After Schrieffer–Wolff elimination of the couplers, the effective qubit-only Hamiltonian contains dispersive shifts $|11\rangle$8 and $|11\rangle$9 such that only the transition $({-}iX)$0 is resonant with the drive on the middle qubit. The ideal target operation is the swap of $({-}iX)$1 and $({-}iX)$2 with a factor $({-}iX)$3, while all other computational states are unchanged (Baker et al., 2021). Numerical evidence gives process fidelity over $({-}iX)$4 with gate time $({-}iX)$5 ns, and a faster operating point of $({-}iX)$6 ns with process fidelity around $({-}iX)$7 (Baker et al., 2021).

In silicon spin qubits, the iToffoli arises from exchange-induced conditional resonance of the middle spin in a linear triple quantum dot. With both nearest-neighbor exchanges turned on, the effective Hamiltonian decomposes into four target-qubit blocks $({-}iX)$8, one for each outer-spin configuration. Choosing the EDSR drive so that only the $({-}iX)$9 sector is resonant makes the native operation a controlled-controlled flip in the $|00\rangle$0 subspace; the paper identifies this native gate as an $|00\rangle$1-Toffoli and then shows how to obtain standard Toffoli by adding a $|00\rangle$2 C-Phase (Gullans et al., 2019). The protocol reports gate times on the order of $|00\rangle$3 ns, with examples such as $|00\rangle$4 ns and ideal infidelity as low as $|00\rangle$5, and the abstract states fidelity exceeding $|00\rangle$6 (Gullans et al., 2019).

A closely related theoretical single-step construction uses one target qubit strongly Ising-coupled to $|00\rangle$7 controls and resonantly driven only in the selected control sector. The resulting primitive acts as $|00\rangle$8 on the target when all controls are $|00\rangle$9, which the paper explicitly calls the $|11\rangle$0-Toffoli (Rasmussen et al., 2019). For the two-control case, simulations with decoherence gave fidelities just above $|11\rangle$1 at $|11\rangle$2, corresponding to $|11\rangle$3 ns, and above $|11\rangle$4 for the ancilla-based conversion to standard Toffoli (Rasmussen et al., 2019).

4. Trapped-ion and topological realizations

In trapped ions, the central object is an $|11\rangle$5-control-qubit $|11\rangle$6-Toffoli implemented natively by selective-subspace inversion in the $|11\rangle$7-basis rather than by decomposition into pairwise entanglers. The effective spin Hamiltonian is

$|11\rangle$8

or, with the transverse control field along $|11\rangle$9,

$CCX$0

Choosing $CCX$1 to cancel the selected control-sector splitting and applying a pulse with $CCX$2 makes the target flip with the extra factor $CCX$3 only in that selected sector (Goel et al., 2021). This is a direct multiqubit interaction episode with no ancilla for the native gate itself. The paper reports that the good-fidelity 2-control gate time is about $CCX$4 ms and concludes that such gates are about an order of magnitude or more slower than two-qubit entangling gates; for larger $CCX$5, desired-flip probabilities remain high, such as $CCX$6 for $CCX$7 and $CCX$8 for $CCX$9 (Goel et al., 2021).

In the Fibonacci anyon model, the iToffoli serves as a benchmark for a controlled-injection compilation method for topological three-qubit gates. The method uses four-anyon encoding, grouped anyon-pair injection, and the gate family $1$0 to realize controlled-controlled-$1$1 operations with $1$2 (Tounsi et al., 2023). A central claim is that controlled-injection uses only four two-qubit gates—$1$3, $1$4, $1$5, and $1$6—rather than the conventional five-two-qubit-gate decomposition, and that in the compiled topological comparison this corresponds to 4 two-qubit gates rather than 7 (Tounsi et al., 2023). For braid length $1$7, the controlled-injection route has total length $1$8 and depth $1$9, compared with $-iX$00 and $-iX$01 for the decomposition approach. The displayed numerical iToffoli of Fig. 8 is reported to represent iToffoli up to $-iX$02 distance error, with leakage of order $-iX$03 (Tounsi et al., 2023).

5. Algorithmic role and compilation advantages

The main algorithmic demonstration of a native iToffoli appears in the quantum computation of frequency-domain molecular response properties for NaH and KH. There the hardware-native three-qubit gate is used as the core primitive for synthesizing CCZ inside LCU circuits for off-diagonal response-function terms (Sun et al., 2023). Under linear superconducting-qubit connectivity, CCZ synthesized only from CZ gates requires 8 CZs, whereas the iToffoli-based route uses the native three-qubit gate to supply the $-iX$04 part and adds a long-range $-iX$05 to cancel the phase factor $-iX$06 (Sun et al., 2023). In the off-diagonal circuits, this reduces depth from $-iX$07–$-iX$08 to $-iX$09–$-iX$10, and reduces entangling overhead from $-iX$11–$-iX$12 native two-qubit gates to $-iX$13 iToffoli gates plus $-iX$14–$-iX$15 native two-qubit gates (Sun et al., 2023).

This depth reduction is the practical reason iToffoli attracts attention. The fixed-frequency superconducting paper notes that a standard Toffoli requires at least five two-qubit gates for fully connected qubits and eight for nearest-neighbor connectivity, whereas a native three-qubit gate can absorb much of that overhead into one calibrated interaction (Kim et al., 2021). In topological compilation, the same logic appears in braid language: reducing the number of logical two-qubit ingredients shortens braid length, reduces depth, and lowers the search burden for braid approximants (Tounsi et al., 2023). A plausible implication is that iToffoli is especially valuable in architectures where nonlocal or high-fidelity two-qubit gates are expensive but one platform-native three-qubit interaction exists.

The gate also plays a role in gate-synthesis theory. In the fixed-frequency superconducting work, numerical synthesis with a fixed three-qubit gate plus arbitrary single-qubit $-iX$16 layers showed that either Toffoli or iToffoli can synthesize arbitrary three-qubit Clifford targets at depth $-iX$17 and sampled Haar-random targets at $-iX$18, while other simultaneously driven three-qubit gates can reach $-iX$19 (Kim et al., 2021). This places iToffoli within a wider family of useful native three-qubit compilation primitives rather than as an isolated gate.

6. Variants, standardization issues, and common misconceptions

A recurring misconception is that iToffoli is just standard Toffoli times a global phase. In the single-qubit sense $-iX$20, but once the factor appears only in the activated control sector, it becomes a conditional phase and therefore changes the multi-qubit gate nontrivially (Goel et al., 2021). This is why some papers must add a controlled $-iX$21, a $-iX$22 C-Phase, or other phase bookkeeping to recover canonical CCX or CCZ behavior (Sun et al., 2023, Gullans et al., 2019).

The term itself is not fully standardized. Some works use control on $-iX$23; others use open controls on $-iX$24 and then convert by local $-iX$25 gates on the controls (Kim et al., 2021). Some define the activated target operation as $-iX$26, others as $-iX$27 (Tounsi et al., 2023, Rasmussen et al., 2019). Some use the gate chiefly as a native building block for CCZ rather than as a final algorithmic primitive (Sun et al., 2023). For this reason, exact specification of control convention, qubit ordering, and local phase frame is essential in both theory and experiment.

A second source of confusion is that many “Toffoli” papers are not about iToffoli at all. The diffractive-neural-network photonic experiment realizes a standard three-qubit Toffoli and explicitly states that it does not define or implement an iToffoli (Wang et al., 2024). The IBM hardware-conscious optimization paper targets exact $-iX$28, not a phase-modified Toffoli (Bowman et al., 2022). The single-shot transmon control paper optimizes a phase-flexible CCZ and then obtains Toffoli by Hadamards, but not an iToffoli (Zahedinejad et al., 2015). Likewise, adiabatic optical, Rydberg, and recent hybrid-code transversality work focus on standard Toffoli or CCZ-family gates rather than iToffoli specifically (Grigoryan et al., 2013, Ashkarin et al., 2021, Jiao et al., 12 Nov 2025).

Within current research, iToffoli is therefore best understood not as a single universally fixed gate, but as a controlled-controlled target flip with a prescribed nontrivial phase convention whose exact form is platform-dependent. What remains stable across the literature is the structural role: native iToffoli implementations reduce decomposition overhead, expose conditional phase control directly at the hardware level, and sit naturally between standard Toffoli, CCZ, and broader phase-twisted three-qubit gate families (Kim et al., 2021, Baker et al., 2021, Sinha et al., 2024).