The stable secondary-to-primary flux ratios of cosmic rays (CRs), represented
by the boron-to-carbon ratio (B/C), are the main probes of the Galactic CR
propagation. However, the fluorine-to-silicon ratio (F/Si) predicted by the CR
diffusion coefficient inferred from B/C is significantly higher than the latest
measurement of AMS-02. This anomaly is commonly attributed to the uncertainties
of the F production cross sections. In this work, we give a careful test to
this interpretation. We consider four different cross-section parametric
models. Each model is constrained by the latest cross-section data. We perform
combined fits to the B/C, F/Si, and cross-section data with the same
propagation framework. Two of the cross-section models have good overall
goodness of fit with ${\chi}^{2}/{n}_{d.o.f.}\sim 1$ . However, the goodness of fit of
the cross-section part is poor with ${\chi}_{\mathrm{c}\mathrm{s}}^{2}/{n}_{\mathrm{c}\mathrm{s}}\gtrsim 2$ for
these models. The best-fitted F production cross sections are systematically
larger than the measurements, while the fitted cross sections for B production
are systematically lower than the measurements. This indicates that the F
anomaly can hardly be interpreted by neither the random errors of the
cross-section measurements nor the differences between the existing
cross-section models. We then propose that the spatially dependent diffusion
model could help to explain B/C and F/Si consistently. In this model, the
average diffusion coefficient of the Ne-Si group is expected to be larger than
that of the C-O group.