Answers are grounded in this paper and cite their sources.
GaN on large, cheap silicon wafers is the industry’s route to affordable power and RF electronics, but GaN and Si differ by 17 % in lattice constant and by a factor of two in thermal expansion. Project Angelica asks one question: can a rare-earth-oxide foundation between Si and the nitride make GaN-on-Si easier to build? The question is followed through a reproducible multi-physics model in which every constant carries a source. A (Gd₀.₅₈Eu₀.₄₂)₂O₃ base is lattice-matched to Si(111) at its 650 °C growth temperature. It was chosen with Angie’s Mixes, an open compound-selection framework. Two composition grades carry the stack upward. Grading-O, in the oxide, is a lattice bridge to a nitrided ScN cap. Grading-N, in the nitride (AlN→GaN), stores compression by threading-dislocation inclination and is jointly calibrated to measured wafers. Fracture, Schrödinger–Poisson transport, 2D thermal, charged 2D breakdown and RF-loss models follow the stack to the device. Differential Evolution, Monte Carlo, slip, dynamic-Ron, electro-thermal and cost models stress-test it. A running scoreboard records what the oxide wins, ties and costs. Main results: A 0.12 µm oxide moves a 650 V buffer inside the stress/bow window. Under shared model uncertainties, Angelica designs pass in 17–30 % of samples, against 9 % for the conventional stack. The 650 V buffer can be 17 % thinner if the oxide carries its DC voltage share. The cost is 3–10 % higher thermal resistance. Every claim is tagged IMPLEMENTABLE, TESTABLE or CONCEPT, and one command regenerates every number. This record contains the formal paper in English, Traditional Chinese and Japanese, a Primer in all three languages, and the full code and sources (manuscript CC BY 4.0, code Apache 2.0). The Reader’s Guide remains in version 1 (DOI 10.5281/zenodo.23806034). This revised edition (October 2026) replaces the first edition (June 2026, DOI 10.5281/zenodo.20608888). All changes are documented in the section “Relation to the first edition”. v2 update (October 2026). Version 2 pushes the applications further, with every new number reproducible from the code: Wafer size and voltage. On 300 mm wafers no conventional single-grade stack of the model fits the stress/bow window. Angelica fits with 0.46–1.35 µm of oxide, even if the oxide blocks only its capacitive share. At 1200 V a conventional stack also fits on 2 mm Si. Angelica moves the 1.5 mm Si case inside with 0.34 µm of oxide, or cuts the stack by 38 % if the oxide carries DC. An industry strain-relief stack already reaches 300 mm, so Angelica is presented as added margin, not as the only route. Monolithic power ICs. The oxide alone does not stop back-gating at switching speed. A source-tied ScN back-plane on the oxide could, for half-bridges up to about 150 V. Other applications. Bidirectional 800 V switches, RF loss converted to power-amplifier output (worst-case bounds), and a bonding rule for transfer to diamond (below about 150–175 °C). The update also covers size rules for released MEMS membranes and thermal-neutron capture in the (Gd,Eu) base: about 4 %, harmless, and a possible built-in detector. Memristors stay a concept. Process changes. Three new go/no-go experiments; the version-2 changes were checked by an independent review before release. Apart from these additions, the version-1 content is unchanged; see “Relation to the first edition”.
No takes yet. Share an insight, caveat, or question.
Tung Ning Liu (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: