← All industry projectsReal-time thermal estimation
LG DISPLAY · 2025

Real-time thermal estimation

From a physical model to a hardware datapath

I developed a pixel-level thermal model that turns video-derived current maps into temperature estimates through 2D heat diffusion. The work adapted the algorithm to real-time display hardware and reached RTL handoff before tape-out.

IndustryLG Display · 2025RTL handoff
Explore the approach ↓
01 / THE PROBLEM

How can a spatial thermal model become a practical real-time datapath?

Panel temperature depends on both the video-dependent electrical load and heat transfer between neighboring pixels. A pixel-wise estimator must capture this spatial behavior while fitting the execution constraints of a real-time display pipeline. The central engineering task was to make a physical thermal model usable as a hardware algorithm.

02 / THE APPROACH

A content-driven spatial thermal model

01

Use current maps as the model input

Represent the load produced by displayed video as a pixel-current map. This connects temperature estimation to the content being driven on the panel.

02

Model heat diffusion across the panel

Develop a pixel-wise prediction algorithm based on two-dimensional heat diffusion. Neighboring locations contribute to the estimated thermal state, capturing the spatial behavior that a purely independent per-pixel estimate would miss.

03

Adapt the computation for real-time hardware

Optimize the algorithm against implementation constraints and carry the model through RTL handoff before tape-out. The work joins physical modeling with the requirements of the display datapath.

03 / MY CONTRIBUTION

Thermal modeling and hardware adaptation

Developed the thermal-prediction algorithm and adapted it to hardware implementation constraints.

04 / OUTCOME & SCOPE

Algorithm completed through RTL handoff

I developed the thermal-prediction algorithm, adapted it to real-time hardware constraints, and completed RTL handoff prior to tape-out. The project demonstrates the conversion of a spatial physical model into an implementable estimation pipeline.