Embedded Photonic Bell Measurements

• Embedded photonic Bell measurements are entangling measurements incorporated into fiber, chip, and frequency-domain architectures to enable fusion in photonic quantum computing and networking.
• They overcome the 50% success limit of passive linear optics by redistributing complexity into ancillae, auxiliary degrees of freedom, and integrated detector schemes.
• Real-world implementations use platforms like fibre-based multiports, silicon circuits, and quantum memories, achieving enhanced discrimination probabilities and process fidelities.

Embedded photonic Bell measurements are Bell-state measurements implemented not as isolated bulk-optics primitives, but as entangling measurements embedded in fiber-compatible hardware, integrated photonic circuits, frequency-domain processors, graph-state fusion networks, stabilizer encodings, or memory-coupled network nodes. In this literature, Bell measurements remain the central joint projection onto the Bell basis, but their operational role is broader: they serve as fusions in fusion-based photonic quantum computing, as swap operations in quantum networking, and as encoded logical readout inside larger photonic resource states. The field is organized by a persistent tension between the standard passive-linear-optics success limit of $50\%$ and a growing set of embedded strategies that redistribute complexity into ancillae, auxiliary degrees of freedom, detector architectures, or encoded resource states (Bianchi et al., 23 Sep 2025, Hauser et al., 2024).

1. Linear-optical foundation and the $50\%$ constraint

For qubits, the Bell basis is

$$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$

In photonic dual rail,

$$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$

so Bell states become two-photon, four-mode states. The standard photonic Bell-state measurement is a linear-optical interferometric analyzer built from a $50{:}50$ beam splitter, possibly supplemented by phase shifters, wave plates, polarizing beam splitters, and single-photon detectors. Its key physical effect is Hong–Ou–Mandel interference: symmetric two-photon amplitudes bunch, while antisymmetric components anti-bunch. In a typical balanced beam splitter,

$$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$

Operationally, only two Bell states produce unique coincidence structures, while the other two remain degenerate, so for dual-rail qubits under passive linear optics one has the canonical bound

$P_s \le 50\%.$

The review literature is explicit that this is not a weakness of a particular setup but a resource-theoretic limit under passive linear optics with no nonlinearities, no ancillary photons, and no extra resources capable of lifting Bell-state degeneracy (Bianchi et al., 23 Sep 2025).

The same limit appears in the photonic-fusion language used for graph states and FBQC. There, a successful fusion is a Bell-state measurement that destructively measures two commuting weight-two Pauli parities on qubits $A$ and $B$. Standard photonic type-II fusion typically has $p_s=0.5$, and on failure it reduces to two single-qubit Pauli measurements. Embedded photonic Bell measurements therefore inherit both the utility and the incompleteness of linear-optical Bell analysis: they are indispensable, but without additional resources they remain partial and probabilistic (Löbl et al., 2024).

2. Embedded hardware realizations

Embedded implementations span fiber-compatible multiports, partially integrated silicon circuits, frequency-domain processors, and hybrid memory-coupled nodes. What changes across platforms is not the Bell-basis definition, but where the nontrivial resource resides: in ancillary photons, in integrated interferometers, in frequency mixing, or in a matter memory.

Platform	Bell functionality	Representative result
Fibre-based $50\%$0 multiport with ancillary Bell pair (Hauser et al., 2024)	Boosted photonic BSM for FBQC	$50\%$1; photon-loss threshold $50\%$2 versus $50\%$3 non-boosted
Silicon quantum photonic circuit with off-chip simplified BSM (Liu et al., 2021)	On-chip generation and switching of $50\%$4; off-chip partial discrimination of $50\%$5	simplified BSM raw visibility $50\%$6
Quantum frequency processor for spectrally distinct photons (Lingaraju et al., 2021)	Partial BSA in frequency space for $50\%$7	accuracies $50\%$8 for $50\%$9 and $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$0 for $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$1
Single-ion quantum memory plus heralded absorption (Arenskötter et al., 2023)	Full Bell-basis measurement of an atom-photon 2-qubit state	mean process fidelity $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$2 uncorrected, $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$3 corrected

The fibre-based boosted analyzer of "Boosted Bell-state measurements for photonic quantum computation" is the clearest example of an embedded photonic BSM in the strict hardware sense. It uses two Sagnac Bell-pair sources, a fused-fiber $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$4 splitter implementing a discrete Fourier transform, polarization-resolved outputs, and pseudo photon-number resolution via $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$5 spatial multiplexing into 56 SNSPDs. The ancillary Bell pair lifts the $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$6 degeneracy so that the theoretical average discrimination probability becomes $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$7, while the experiment reports $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$8, limited mainly by higher-order SPDC emissions, residual distinguishability, detection nonidealities, and imperfect pseudo-PNR (Hauser et al., 2024).

The silicon circuit of "Generation and dynamical manipulation of polarization entangled Bell states by a silicon quantum photonic circuit" is embedded only in a narrower sense. The chip integrates pump splitting, SFWM pair generation in four $$|\Phi^\pm\rangle = \frac{1}{\sqrt{2}\left(|00\rangle \pm |11\rangle\right)}, \qquad |\Psi^\pm\rangle = \frac{1}{\sqrt{2}\left(|01\rangle \pm |10\rangle\right)}.$$9 silicon waveguides, coherent superposition, phase control, and path-to-polarization conversion, but the Bell measurement itself is off-chip: a $$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$0 fiber coupler, variable delay, polarization control, and APD coincidence counting implement a simplified BSM that distinguishes $$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$1 from $$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$2. The work is therefore an integrated Bell-state source and manipulator with off-chip partial Bell-state discrimination, not a full on-chip Bell analyzer (Liu et al., 2021).

The frequency-domain analyzer of "Bell state analyzer for spectrally distinct photons" removes a different bottleneck: it performs Bell-state discrimination without first forcing the photons to be spectrally identical. A quantum frequency processor built from EOM $$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$3 pulse shaper $$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$4 EOM implements interleaved Hadamard gates in discrete frequency bins, converting the sign in

$$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$5

into different coincidence patterns in frequency space. It remains a partial analyzer—$$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$6 are unresolved—but it directly addresses wavelength-multiplexed and heterogeneous-node settings in which spectral mismatch is intrinsic rather than parasitic (Lingaraju et al., 2021).

The single-ion memory experiment occupies a broader category of embedded Bell measurement. It is not a purely photonic integrated analyzer, but it does embed Bell-basis readout into a network node containing a quantum memory. One photonic qubit is mapped to the memory, heralded absorption plus Raman emission realize the state-selective interaction, and the combination of herald-photon polarization, passage number, and final atomic readout resolves all four Bell states of an atom-photon pair (Arenskötter et al., 2023).

3. Embedded resources beyond passive linear optics

The main routes beyond the passive-linear-optics limit are resource upgrades. One family uses ancillary photonic states. In Grice’s scheme, supplementing the unknown Bell state with $$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$7 entangled auxiliary states yields

$$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$8

so $$|0_L\rangle = |10\rangle,\qquad |1_L\rangle = |01\rangle,$$9 gives $50{:}50$0. Ewert and van Loock replace those ancillae with simpler auxiliary states built from two identical single photons interfering on a beam splitter, with

$50{:}50$1

A different route uses predetection squeezing, reaching

$50{:}50$2

The review treats ancilla-assisted linear optics, squeezing, nonlinear optics, hyperentanglement, and cavity-QED/light–matter interfaces as the principal strategies for surpassing $50{:}50$3 (Bianchi et al., 23 Sep 2025).

A more specifically embedded variant is ancilla-photon-free assistance by additional degrees of freedom on the same two photons. "Auxiliary Schmidt Rank as a Resource for Photonic Bell Measurements" proves that for two system qudits of local dimension $50{:}50$4 with a fixed auxiliary entangled state $50{:}50$5 carried by the same photons, a single conclusive Bell-label functional exists iff

$50{:}50$6

while deterministic discrimination of all $50{:}50$7 Bell labels requires

$50{:}50$8

A maximally entangled rank-$50{:}50$9 auxiliary state saturates the deterministic bound via local Bell-basis sorting between each photon’s system and auxiliary degrees of freedom. In this formulation, auxiliary Schmidt rank—not merely larger mode space—is the certified embedded resource (Laha et al., 23 Jun 2026).

Logical encodings offer another embedded route. In "Nearly Deterministic Bell Measurement for Multiphoton Qubits and Its Application to Quantum Information Processing", one logical qubit is encoded into an $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$0-photon GHZ-type state and measured via $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$1 standard physical Bell measurements, each still $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$2-efficient. The logical Bell measurement succeeds unless all $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$3 elementary BMs fail, giving

$$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$4

This uses only linear optics and photon on-off detectors and does not require photon-number-resolving detection, but it shifts the resource burden into multiphoton GHZ-state generation and loss management (Lee et al., 2015).

At the level of stabilizer encodings, "Optimal logical Bell measurements on stabilizer codes with linear optics" proves that any destructive logical BM on two stabilizer-encoded logical qubits can be mapped onto a single physical BM on an arbitrary qubit pair from the two codes. The resulting general upper bound is

$$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$5

For standard linear optics, $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$6, so identical $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$7-qubit code blocks satisfy $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$8. The paper then gives sufficient criteria and explicit schemes attaining this bound for quantum parity, five-qubit, planar and rotated surface, tree, and seven-qubit Steane codes (Reiß et al., 13 Jan 2026).

4. Bell measurements embedded in graph states and FBQC

In graph-state photonics, Bell measurements appear as fusions. "Transforming graph states via Bell state measurements" formalizes a successful fusion as the simultaneous measurement of two commuting weight-two Pauli parities on qubits $$\hat a^\dagger \rightarrow \frac{\hat c^\dagger + \hat d^\dagger}{\sqrt{2}}, \qquad \hat b^\dagger \rightarrow \frac{\hat c^\dagger - \hat d^\dagger}{\sqrt{2}}.$$9 and $P_s \le 50\%.$0 already embedded in a larger graph state. Rather than treating a Bell measurement as a two-qubit object in isolation, the paper derives the induced transformation on the entire graph. It identifies five distinct types of fusion success cases and gives a full set of graph-transformation rules, including connected fusion qubits and overlapping neighborhoods—precisely the regimes that arise in realistic fusion networks after earlier successful fusions have modified connectivity (Löbl et al., 2024).

The standard Bell/fusion type $P_s \le 50\%.$1 is only one of these cases. Rotated Bell analyzers generated by local Clifford preprocessing realize additional parity pairs such as $P_s \le 50\%.$2, $P_s \le 50\%.$3, $P_s \le 50\%.$4, and $P_s \le 50\%.$5. The resulting graph updates are expressed in terms of symmetric differences of neighborhoods, which the paper visualizes by Venn diagrams. This is the graph-theoretic core of embedded Bell measurement: the measured qubits are destroyed, but the rest of the resource graph is rewired in a precisely characterizable way (Löbl et al., 2024).

The FBQC relevance becomes explicit in the boosted-fusion experiment. There, the BSM is not evaluated only as an optical component; its measured success probability is propagated into a six-ring fusion network using $P_s \le 50\%.$6-Shor encoded six-ring resource states. Encoded fusion is chosen so that

$P_s \le 50\%.$7

and the threshold simulations with Plaquette and FusionBlossom show that the experimentally realized boosted BSM with

$P_s \le 50\%.$8

supports a tolerable individual-photon loss probability of $P_s \le 50\%.$9, whereas the non-boosted case with

$A$0

supports only $A$1. The architectural lesson is explicit: embedded Bell measurements should be judged by how component-level success probabilities translate into loss thresholds and syndrome-graph connectivity in large fusion networks (Hauser et al., 2024).

5. Detector engineering, networking, and distributed Bell measurements

Practical embedded Bell measurements are frequently limited less by formal Bell-basis structure than by detector behavior and network nonidealities. "A Hybrid Approach to Mitigate Errors in Linear Photonic Bell-State Measurement for Quantum Interconnects" addresses false heralding from multi-photon components when detectors cannot resolve photon number. Its hybrid Bell-state measurement accepts an event only when an on-off SPD clicks and a homodyne quadrature outcome lies within a window of width $A$2 around $A$3. In the small-$A$4, small-$A$5 limit,

$A$6

so the two-photon false-herald term is suppressed by homodyne efficiency. The corresponding truncated POVM purity tends to $A$7 as $A$8. At the protocol level, the paper reports that HBSM with $A$9 yields higher teleportation fidelity than ideal PNR up to $B$0, and higher swapping fidelity up to $B$1 (Asenbeck et al., 2024).

A complementary systems comparison appears in "Entangling Quantum Memories via Heralded Photonic Bell Measurement". There the Bell measurement is embedded at a midpoint relay connecting two emissive memories. The analysis compares single-rail and dual-rail photonic encodings under loss, imperfect swap visibility, stochastic carrier-phase mismatch, and detector excess noise. Single rail has the better high-loss asymptote, with success scaling as $B$2, whereas dual rail scales as $B$3. But dual rail outperforms single rail in low-loss links, is unaffected by stochastic carrier-phase mismatch, and is less severely affected by excess noise. The rate metric is

$B$4

with $B$5 the hashing bound lower bound to distillable entanglement. The paper explicitly identifies excess noise as a finite-range limiter because for any $B$6, the distillable entanglement drops to zero beyond a finite total channel-loss threshold (Dhara et al., 2023).

A more distributed rethinking of topology appears in "Informationally symmetrical Bell state preparation and measurement". There, Bell discrimination is implemented without requiring the two photons to overlap at a common beam splitter. The scheme remains a passive-linear-optical partial Bell analyzer with

$B$7

but it is symmetrically divisible between two parties and identifies $B$8 rather than the usual $B$9 sector. The reported visibilities are $p_s=0.5$0, $p_s=0.5$1, and $p_s=0.5$2 in the relevant interference signatures. For embedded photonics, the importance is architectural: centralized HOM overlap is not conceptually necessary, though the requirement is effectively shifted into distributed phase-stable amplitude exchange (Kim et al., 2018).

6. Conceptual boundaries and open problems

Several recurrent misconceptions are corrected by the current literature. First, an integrated Bell-state source or Bell-state manipulator is not yet a full embedded Bell analyzer: the silicon circuit of (Liu et al., 2021) explicitly demonstrates on-chip generation and switching between $p_s=0.5$3 and $p_s=0.5$4, but the Bell measurement remains off-chip and partial. Second, additional modes alone do not certify stronger Bell measurement; the exact resource threshold in the ancilla-photon-free same-photon setting is auxiliary Schmidt rank $p_s=0.5$5, not mere modal enlargement (Laha et al., 23 Jun 2026). Third, improving component-level success probability is not automatically the right figure of merit; the FBQC threshold analysis shows that the decisive question is how Bell measurement statistics propagate into fusion-network connectivity and loss tolerance (Hauser et al., 2024).

Open challenges remain layered. At the fundamental level, passive linear optics without extra resources is still capped at $p_s=0.5$6 for dual-rail qubit BSM, while high-dimensional Bell analysis faces even stronger no-go restrictions. At the engineering level, the review identifies deterministic high-quality ancilla generation, low-loss and high-efficiency nonlinear interactions, reliable PNRDs, detector dark counts, phase instability, and faithful scaling to larger resource states or higher qudit dimension as the main bottlenecks. It also notes that path encoding maps especially well to integrated photonics, making path-encoded interferometric BSMs, ancilla-assisted linear-optics analyzers with integrated sources and detectors, and integrated nonlinear waveguide approaches the most natural chip-compatible directions (Bianchi et al., 23 Sep 2025).

Taken together, the literature defines embedded photonic Bell measurements less by one device geometry than by one structural idea: Bell projection is executed inside a larger photonic system—telecom fiber hardware, an integrated interferometric circuit, a frequency processor, a graph-state resource, a stabilizer code, or a memory-coupled node—and its usefulness is determined jointly by measurement completeness, resource embedding, detector realism, and the architecture-level task it serves.