Presentazioni e Articoli Tecnici

The transition to glass-core PCB substrates is driven largely by electrical advantages — low dielectric loss, CTE match to silicon. The glass substrate itself is a substantially better thermal conductor compared to FR-4 epoxy (kglass ≈ 1.4 W/m·K vs. kFR4 ≈ 0.3 W/m·K), producing broader lateral heat spreading, more pronounced thermal crosstalk, and faster transient response than FR-4 at equivalent power. Established design rules from IPC-2152, developed for FR-4, do not account for this difference and cannot be directly transferred to glass-substrate layouts.

Existing thermal studies of glass substrates focus on vertical heat flow and component-level packaging. This work characterizes lateral heat spreading, inter-component crosstalk as a function of separation distance, and transient thermal time constants under pulsed loading — the system-level thermal behavior that governs component placement in dense layouts. The finding that glass conducts heat laterally better than FR-4 epoxy is counterintuitive and has immediate practical implications for anyone placing power components on glass PCBs. The COMSOL Joule Heating multiphysics interface couples the Electric Currents interface (AC/DC Module) with Heat Transfer in Solids (Heat Transfer Module) via the Electromagnetic Heating coupling node. Two SMD resistive heaters on the substrate are driven at controlled current levels; the model self-consistently computes resistive power dissipation and three-dimensional thermal conduction with convective surface boundary conditions. Stationary studies parametrically sweep heater separation from 5 to 30 mm, extracting crosstalk magnitude at the midpoint between heaters as a function of distance. Time-dependent studies apply step inputs and extract thermal time constants, characterizing how each substrate stores and releases heat under pulsed loading representative of AI accelerator workloads.

Simulation results show glass reaching steady-state peak temperatures approximately 40 K above ambient at 0.8 W heater input, with thermal halos extending broadly and measurable crosstalk persisting at large separations. FR-4 at identical power reaches approximately 85 K above ambient — more than twice the peak temperature — but with a sharply concentrated thermal field and crosstalk that decays more rapidly with distance. Transient simulations confirm glass has approximately six times higher thermal diffusivity than FR-4 epoxy, producing faster rise-and-fall response under pulsed inputs, with direct implications for pulsed-load thermal budgets in AI packaging.

A purpose-built instrumented PCB provides experimental ground-truth for every model prediction. Built around an STM32F411 microcontroller, the board integrates two WSL2512 precision SMD power resistors as heaters, each driven through an AO3400A N-channel MOSFET that enables software-programmable DC and pulsed current profiles. Heater current is measured in real time via KRL3216 shunt resistors and INA181 current-sense amplifiers, providing independently verified input power per channel. Twelve NTCG163JF103FT1 NTC thermistors distributed across the substrate map the complete spatial thermal field, directly measuring the crosstalk gradients the COMSOL model predicts. An MCP79410 hardware RTC with coin-cell battery provides accurate timestamps, and a microSD card stores full thermal logs for post-processing comparison against simulation. The validated framework contributes the first quantitative, substrate-specific thermal design guidance for glass PCBs, applicable to AI accelerator packaging, RF front-end modules, and high-density power electronics evaluating glass as a substrate platform.