Project Proteus: One Nitride, Three Forms, Three Puzzles

Project Proteus follows the Angelica series onto the nitrides. Near lattice match to GaN, one wurtzite ScₓAl₁₋ₓN layer can be a transistor barrier, a ferroelectric switch and a piezoelectric resonator, and each form hides a disagreement in the literature. On the Angelica GaN-on-Si foundation, the paper computes everything that can be computed before a wafer is needed, and ends with five test vehicles for what computation cannot settle. Every result is tagged IMPLEMENTABLE, TESTABLE or CONCEPT. Puzzle 1, the matched composition: laboratories report 0.09, 0.11 and 0.14; 0.18 is the theoretical value. No common match with composition-scale errors of 7–10 % reconciles the three measurements (χ² = 13.8, 2 degrees of freedom).Puzzle 2, the missing mobility: with Angelica's calibrated scattering model, two laboratories' ScAlN HEMTs need an extra scattering channel of about 641–658 cm²/Vs. ScAlN buys current (5.7–8.1 A/mm, 2.5–3.5× AlGaN), not speed.Puzzle 3, two ferroelectrics: sputtered films switch at 4–5.5 MV/cm, epitaxial films near match at 0.65–1.2 MV/cm. In a HEMT gate stack, 2DEG screening and interlayer breakdown let only about 27–45 % (epitaxial) and 2.5 % (sputtered) of the polarization switch, enough for the reported 1–2 V windows.Form 3: a 39 GHz (n260) resonator needs x ≥ 0.175 under an optimistic bandwidth rule and x ≈ 0.30 under a conservative one, so the matched barrier composition is at best marginal for it.On silicon, heat, not current, caps power (4.0 vs 8.8 W/mm on SiC for +175 K); the RF loss of the stack depends on whether a parasitic layer forms at the silicon surface, not on the oxide thickness. Formal paper in English (authoritative), Traditional Chinese and Japanese; primers for middle-school readers in the same three languages; full code and sources. Not peer-reviewed. Defensive publication. Text CC BY 4.0, code Apache 2.0.