Back around 1980, the GaAs technology suddenly challenged silicon for speeding-up circuitry in the SSI, MSI, and LSI architecture complexity ... VLSI being still CMOS silicon and slower. [Several improved silicon technologies were being considered, including I2L or IIL Integrated Injection Logic, and FACT and FAST [Fairchild: brand] Advanced CMOS and Schottky Technology(s) utilizing fast low-power internal logic hybrid-driving bipolar line-levels to the pins]
But GaAs presents a nuclear TEMPEST challenge: to wit: both gallium and arsenic are susceptible to nuclear neutron-doping, and radio-decaying in the same atomic-weight direction: both move up to the next element in about a month - thus changing the p-n doping character within a ten-atom radius region of the substrate: In GaAs this is important because intrinsic GaAs is used to provide narrow [high-speed] ballistic-electron-wells, and any slight modification by neutrons may create new paths through the substrate. VHDL Very-High-Density-Logic for TEMPEST-engineering applications may achieve billions of potential single-point-failures in neutron-intense nuclear and space environments - and GaAs must be restricted.
Undoing the effects of neutron-doping of GaAs is equally challenging: pure germanium of one fairly rare isotope may be added to the GaAs to statistically compensate that drift by radio-decaying oppositely, but with 53% ultra-pure isotope proportion, this is cost-prohibitive.
Comparatively, silicon and carbon, in silicon-carbide substrates, both have abundant isotopes which absorb neutrons without effect. The band-gap and thermal stability of silicon-carbide is higher than for silicon alone: high-enough to approach GaAs speed.