XCOM: Multifaceted Technical Constructs

• XCOM is a term representing multiple, domain-specific constructs, including a photon cross-section database for radiation shielding and a synchronization layer for quantum control systems.
• In radiation shielding, XCOM benchmarks mass attenuation coefficients using standard methods such as the Lambert–Beer law and mixture rules, ensuring simulation validation.
• Variants like miniXCOM and XCom extend the name to game AI and explainable opinion mining, while XCoM in biomechanics represents the Extrapolated Center of Mass.

XCOM denotes several unrelated constructs across contemporary technical literature. In the arXiv record represented here, the term most often refers to a photon cross-section database/program used to compute or benchmark mass attenuation coefficients in radiation-shielding studies, but it also names a full mesh synchronization and communication fabric for QICK-based quantum control systems. Closely related orthographies extend the label into other domains: miniXCOM is a simplified XCOM-inspired tactical environment for game-AI experiments, XCom denotes both an explainable comparative-opinion-mining model and a Cross Completion module in vertical federated learning, and XCoM denotes the Extrapolated Center of Mass in biomechanics and humanoid locomotion (Asadia et al., 2021, Al-Buriahi, 2020, Hamad et al., 2020, Martin et al., 19 Mar 2026, Saadat et al., 2022, Le et al., 1 Mar 2026, Yao et al., 7 Aug 2025, Chinaglia et al., 24 Feb 2026, Huang et al., 28 Aug 2025).

1. Nomenclature and scholarly scope

The same letter sequence appears in multiple literatures, but its meaning is domain-specific and often case-sensitive.

Form	Domain	Meaning
XCOM	radiation shielding	photon cross-section database/program
XCOM	quantum control	synchronization and low-latency communication network for QICK
miniXCOM	game AI	simplified XCOM-inspired tactical combat environment
XCom	NLP	explainable comparative opinion mining model
XCom	VFL	Cross Completion module in X-VFL
XCoM	biomechanics and locomotion	Extrapolated Center of Mass

In radiation-shielding papers, XCOM functions as a computational or benchmark source for the mass attenuation coefficient $(\mu/\rho)$. In quantum-control hardware, XCOM is instead a deterministic inter-board timing and communication layer. In AI, the orthographic variants miniXCOM and XCom denote a tactical test environment and two unrelated algorithmic systems. In locomotion and gait analysis, the visually similar but distinct symbol XCoM refers to Extrapolated Center of Mass rather than to any XCOM database or network (Asadia et al., 2021, Martin et al., 19 Mar 2026, Saadat et al., 2022, Le et al., 1 Mar 2026, Yao et al., 7 Aug 2025, Chinaglia et al., 24 Feb 2026).

2. XCOM as a photon cross-section database/program

In radiation-shielding studies, XCOM is used as a trusted source of photon attenuation data for compounds and mixtures. The central quantity is the mass attenuation coefficient, written as $\mu_m$ or $\mu/\rho$. One study makes the attenuation formalism explicit through Lambert–Beer’s law,

$I = I_0 e^{-\mu x},$

with $I_0$ the incident intensity, $I$ the transmitted intensity, $x$ the sample thickness, and $\mu$ the linear attenuation coefficient in $\mathrm{cm^{-1}}$. The associated mass attenuation coefficient is written as

$$\frac{\mu}{\rho} = \mu_m = \frac{1}{\rho x}\ln\!\left(\frac{I_0}{I}\right),$$

and the standard derived shielding quantities are

$\mu_m$0

In that workflow, XCOM is not used to generate every shielding metric directly; it is used to validate the attenuation coefficients before downstream quantities are derived from simulation (Asadia et al., 2021).

A second study describes XCOM as a photon cross-section database/program for multicomponent borate glasses and uses the standard mixture rule,

$\mu_m$1

where $\mu_m$2 is the weight fraction of the $\mu_m$3-th constituent element. From this XCOM-derived $\mu_m$4, the linear attenuation coefficient is obtained from

$\mu_m$5

followed again by

$\mu_m$6

This use is explicitly composition-driven: elemental fractions are supplied to XCOM, and density is then used to recover linear shielding parameters (Al-Buriahi, 2020).

A third study uses XCOM together with Phy-X as a reference for MCNP5-computed mass attenuation coefficients of orthorhombic perovskite ceramics. There the same relations,

$\mu_m$7

organize the photon-shielding analysis. XCOM therefore appears in these papers less as a standalone shielding framework than as the canonical source of $\mu_m$8 against which simulations and derived metrics are anchored (Hamad et al., 2020).

3. Validation workflows and quantitative shielding results

A recurrent XCOM usage pattern is benchmark-first shielding analysis. For the bismuth borate zinc-lithium glasses $\mu_m$9, with $\mu/\rho$0, MCNPX-computed $\mu/\rho$1 values were compared against XCOM over 200 keV to 1500 keV at 200, 400, 800, 1000, and 1500 keV. The abstract reports a relative deviation of 2% between MCNPX and XCOM. The fuller reported RPE ranges were B1: $\mu/\rho$2, B2: $\mu/\rho$3, B3: $\mu/\rho$4, and B4: $\mu/\rho$5, with the conclusion stating that the RPE was less than 4% for all energies except 200 keV. For sample B1, the reported correlation coefficient between MCNPX and XCOM was $\mu/\rho$6. The paper states that B4 provides the best shielding effect, but it also reports MFP and HVL in the order B4 > B3 > B2 > B1 and notes that shielding parameters are inversely related to density. This suggests an internal inconsistency between the stated best-performing sample and the attenuation trend implied by the formulas and densities (Asadia et al., 2021).

For borate-based glasses of the form $\mu/\rho$7, the XCOM energy range is explicitly $\mu/\rho$8 to $\mu/\rho$9, while buildup-factor calculations extend to $I = I_0 e^{-\mu x},$0 and $I = I_0 e^{-\mu x},$1. XCOM and Geant4 are described as being in good agreement overall and “almost overlapping,” but the maximum discrepancies reach about 15%, specifically around $I = I_0 e^{-\mu x},$2 and $I = I_0 e^{-\mu x},$3. The compositional trend is monotone: increasing $I = I_0 e^{-\mu x},$4 increases $I = I_0 e^{-\mu x},$5, decreases HVL and MFP, and raises $I = I_0 e^{-\mu x},$6, so sample F is identified as the best gamma shield, whereas sample D is reported as the best fast-neutron shield with neutron removal cross sections in the range $I = I_0 e^{-\mu x},$7 to $I = I_0 e^{-\mu x},$8 (Al-Buriahi, 2020).

For $I = I_0 e^{-\mu x},$9 ceramics, XCOM was used both at the benchmark energies 0.1, 0.6, 1.25, 5, and 15 MeV and across a broader plotted range of 0.015 to 15 MeV. The reported deviation between MCNP5 and XCOM was less than 6% for all samples and energies, with ranges C1: $I_0$0, C2: $I_0$1, C3: $I_0$2, and C4: $I_0$3. At $I_0$4, the XCOM MAC values rise from 0.9447 for C1 to 0.9559 $I_0$5 for C4; at $I_0$6, they rise from 0.0318 to 0.0323 $I_0$7. The study therefore treats Ni substitution as producing a slight but systematic increase in attenuation, with C4 the best photon attenuator among the four compositions (Hamad et al., 2020).

4. XCOM as a synchronization and communication layer for QICK

In quantum-control hardware, XCOM is a fundamentally different object: it is the inter-board synchronization and low-latency communication fabric introduced for QICK, the Quantum Instrumentation Control Kit. Its stated purpose is to turn multiple QICK boards into a single time-coherent, deterministic control system. The abstract reports synchronization of QICK boards and the absolute clocks governing quantum program execution to within 100 ps, free of drift and loss of lock, together with deterministic all-to-all simultaneous data communication with latency below 185 ns. In the body text, the measured prototype latency for a 32-bit word is reported as 186 ns, with a projected reduction to 62 ns at higher link clocking (Martin et al., 19 Mar 2026).

Architecturally, this XCOM is a full mesh network rather than a packet-switched fabric. If there are $I_0$8 boards, there are $I_0$9 parallel transmit channels, one per board, and every board receives every channel. The current prototype uses a small transceiver card mounted on the FMC connector of each AMD ZCU216 RFSoC board and an external fanout hub that replicates each board’s outgoing LVDS transmit pair to the receive inputs of all boards. Prototype hardware supports up to five RFSoC boards, while the firmware IP already supports up to 15 boards. A notable design choice is the use of a separate forwarded clock for each data channel rather than embedded clock recovery, explicitly to minimize FPGA logic and reduce latency.

The broader timing stack combines an external reference, TI LMK04828B PLL locking in nested zero-delay mode, RFSoC multi-tile synchronization, and XCOM-based alignment of each board’s 48-bit absolute clock counter. During setup, a master board broadcasts RESET and START absolute-clock commands, after which all tProcs run against the same absolute experiment time. The paper reports that the absolute clock wraps around after approximately 8 days without interrupting experiments.

Experimental validation is unusually concrete. Three RF waveforms generated by three different QICK boards and one PMOD digital I/O signal were triggered simultaneously; the measured RF timing skew between the three boards was only 20 ps. The prototype remained synchronized and locked over multiple days, and 100K messages exchanged over loopback paths involving two and three boards exhibited identical latency, which the authors treat as evidence of deterministic long-term latency. The main engineering trade-off is also explicit: full mesh via fanout improves determinism and latency, but wiring density grows with system size.

5. Case-sensitive derivatives in AI and federated learning

miniXCOM is not a study of the commercial XCOM games directly, but a simplified XCOM-inspired tactical combat environment used to evaluate adaptive search. It preserves turn-based grid battle, line-of-sight shooting, squad combat, and positional maneuvering, while omitting broader campaign and class-mechanics structure. The reported environment is a 6 by 6 grid with 2 squad members per side; a unit can move by a maximum of three grid cells; a draw is declared after 20 moves; and if there is line-of-sight, shooting immediately kills the enemy. On this environment, the paper evaluates MCTS-TD, which augments MCTS with temporal-difference learning without pre-training. Over 20 runs of 50 rounds each, with first-player balance enforced by 25 rounds per side moving first, MCTS-TD achieved mean wins per 10 rounds of 6.67 (1.6) against RB1, 6.11 (1.33) against SARSA-UCT, and 5.55 (1.57) against vanilla MCTS, with all pairwise differences reported as significant at the 99% confidence level with $I$0 (Saadat et al., 2022).

XCom in comparative opinion mining is an explainable transformer-based framework for implicit same-user review comparison. The task is to determine whether one reviewed product is better, worse, or similar/non-comparable to another with respect to the aspects appearance, aroma, palate, and taste. The architecture has three phases—aspect-based preprocessing, comparative opinion classification, and prediction explanation—and two principal predictive modules: a score-based branch using aspect-wise rating prediction and a semantic branch using direct textual comparison. The two branches are fused at the probability level,

$I$1

followed by $I$2 classification, and SHAP is used for token-level explanations. On the SUDO beer-review dataset, XCom reports Micro F1 $I$3 and Macro F1 $I$4, compared with Micro F1 $I$5 and Macro F1 $I$6 for the strongest listed baseline, Finetuned-T5 (Le et al., 1 Mar 2026).

A second XCom appears in vertical federated learning, where the term stands for Cross Completion within the X-VFL framework. Here the problem is non-aligned samples with partially missing features and the need for locally independent inference. In the two-client case, XCom reconstructs a client’s missing features from another client’s embedding,

$I$7

followed by re-embedding,

$I$8

This module is coupled to Decision Subspace Alignment, and the paper gives convergence rates of $I$9 for SGD-type algorithms and $x$0 for PAGE-type algorithms for the full training objective. The abstract reports that X-VFL achieves a 15% improvement in accuracy on CIFAR-10 and a 43% improvement on MIMIC-III (Yao et al., 7 Aug 2025).

6. Distinction from XCoM in biomechanics and humanoid locomotion

The orthographically similar XCoM is a separate term meaning Extrapolated Center of Mass. In the markerless 3D TUG pipeline tugturn.py, XCoM is presented as part of the dynamic stability analysis, but the manuscript does not print the XCoM equation, does not define $x$1, and does not specify whether XCoM is computed in 2D or 3D. The concrete XCoM output explicitly reported is XCoM deviation (first gait/second gait): 0.027 / 0.041 m (Chinaglia et al., 24 Feb 2026).

A humanoid beam-walking study gives the explicit XCoM definition under a constant-height LIPM: $x$2 There XCoM supplies the nominal, interpretable footstep template in a two-stage framework: a Stage-1 low-level tracker is trained to robustly follow template footsteps plus perturbations, and a Stage-2 planner predicts a bounded body-frame residual $x$3 for the swing foot only. The reported simulation comparison isolates the role of the template: Template-only: success rate $x$4, Ours (Template + Residual): success rate $x$5, Monolithic RL: $x$6. On hardware, the system is reported to reliably traverse a 0.2 m-wide, 3 m-long beam on a Unitree G1 (Huang et al., 28 Aug 2025).

This suggests that bibliographic disambiguation by spelling alone is unreliable: XCOM, XCom, miniXCOM, and XCoM label unrelated technical objects whose meanings are determined by domain context rather than by the letter sequence itself.