TRL 3.5 Engine • 1024×1024 GPU Mesh • PE-ALD AlN/TiN Interlayer

GPU-Accelerated Thermal Management for GaN-on-Diamond Semiconductors

Closing the 30% thermal error gap of Fourier diffusion models in sub-micron HEMT hotspots. Real-time GPU solver architecture combined with low-TBR PE-ALD interfacial engineering.

1,048,576

GPU Mesh Nodes (1024×1024)

Jacobi-CG

Preconditioned Sparse Solver

≤ 1.5×10⁻⁸

Target TBR (m²K/W)

3.5nm / 1.0nm

PE-ALD AlN/TiN Nucleation

PRODUCT A • SOFTWARE EDA IP

GPU Nanoscale Thermal Physics Engine

Proprietary GPU-accelerated Jacobi-preconditioned Conjugate Gradient (Jacobi-CG) solver with non-linear Picard loops for kappa(T) degradation. Bypasses sparse matrix memory fill-in limitations to deliver sub-micron temperature profile accuracy near the HEMT gate hotspot.

• Zero fill-in Jacobi-CG solver running on CuPy/CUDA

• Multi-finger HEMT channel crosstalk BC support

• Robin Boundary condition interfacial jump abstraction

PRODUCT B • HARDWARE PROCESS IP

PE-ALD Sub-10nm Interlayer Recipe

Graduated Plasma-Enhanced Atomic Layer Deposition (PE-ALD) AlN/TiN (3.5 nm / 1.0 nm) nucleation sequence protecting the GaN buffer from plasma degradation while matching acoustic impedance to CVD Diamond.

• Prevents GaN surface amorphization & oxidation

• Characterization via TDTR Laser Metrology

• Fabless Licensing & Royalty Model

lgdcf_jacobi_cg_bte_solver.py

1M+ Node GPU Mesh

# LGDCF Non-linear Jacobi-CG Thermal Solver (1024x1024 Grid)
import cupy as cp

def solve_jacobi_cg(A_sparse, b, M_inv, x0, max_iter=500, tol=1e-5):
    """Jacobi-preconditioned CG solver bypassing CSR memory fill-in wall"""
    x = x0.copy()
    r = b - A_sparse.dot(x)
    z = M_inv * r
    p = z.copy()
    rs_old = cp.dot(r, z)
    
    for i in range(max_iter):
        Ap = A_sparse.dot(p)
        alpha = rs_old / cp.dot(p, Ap)
        x += alpha * p
        r -= alpha * Ap
        z = M_inv * r
        rs_new = cp.dot(r, z)
        if cp.sqrt(rs_new) < tol:
            break
        p = z + (rs_new / rs_old) * p
        rs_old = rs_new
    return x, i

// PHYSICAL STACK ARCHITECTURE

Heterogeneous Material Stack Integration

GaN Channel (High-Power HEMT)

→

PE-ALD AlN/TiN (Sub-10nm Interlayer)

→

CVD Diamond Substrate

// TARGET MARKETS & APPLICATIONS

Commercial Integration Focus

DEFENSE & RADAR

AESA Transceivers

High power-density GaN HEMT modules.

AEROSPACE

SatCom Amplifiers

Ka/Ku band thermal management.

AI HARDWARE

Power Delivery ICs

Sub-10nm package heat extraction.

METROLOGY

TDTR Validation

Optical thermal characterization.