Burc Gokden | Fromthesky Research Labs

Background

About

Dr. Burc Gokden is an engineering professional and research scientist whose career spans a uniquely diverse set of expertise in applied sciences — from artificial intelligence and large language models to advanced semiconductor manufacturing of microprocessors and solid-state photonics devices.

He is the inventor of the Large Language Model from Power Law Decoder Representations (PLDR-LLM) and the Power Law Graph Transformer (PLGT) — novel architectures that leverage learnable non-linear power law attention mechanisms as a generalization of scaled dot-product attention.

His most recent research demonstrates that PLDR-LLMs exhibit reasoning at self-organized criticality, providing a physics-grounded explanation of how intelligence emerges in large language models.

Prior to founding Fromthesky Research Labs, Dr. Gokden was a Senior ML Engineer at Siemens Digital Industries, a Process Integration and EDA software engineer at Intel (10 nm & 7 nm nodes), and a Postdoctoral Fellow at Harvard University in the Capasso photonics group. He holds a PhD from Northwestern University (2011) specializing in Quantum Cascade Lasers.

Artificial Intelligence & ML

Large Language Models Transformers Non-Linear Attention PyTorch TensorFlow / Keras HuggingFace Graph Neural Networks Decision Trees Shapley Values Generative AI

Semiconductor & Photonics

10 nm / 7 nm Process Integration EDA / Physical Verification Quantum Cascade Lasers Photonic Crystals Cleanroom Fabrication Defect Analysis

Programming & Tools

Python SQL / R / JMP scikit-learn Bash / Perl / Tcl ICV / PXL / StarRC

Focus Areas

Research

⚡

PLDR-LLM Architecture

Novel decoder-only language model architecture using non-linear, multi-head power law graph attention (PLGA) as a generalization of scaled dot-product attention. Open-source pretrained models available on HuggingFace.

🧠

Self-Organized Criticality in LLMs

PLDR-LLMs pretrained at self-organized criticality exhibit reasoning capabilities analogous to second-order phase transitions — providing a physics-grounded explanation of intelligence emergence in large language models.

🔢

Generalizable Tensor Operators

Demonstrated that PLDR-LLMs learn a singularity condition (rank-1 zero-determinant matrix) that acts as a generalizable tensor operator, enabling it to replace its own deep neural net at inference for improved efficiency and interpretability.

🌐

Power Law Graph Transformer

Encoder-decoder transformer for machine translation using learnable power law attention coefficients — demonstrating effective training for NLP sequence-to-sequence tasks.

🔬

Graph Attention Interpretability

CoulGAT: an attention mechanism inspired by the screened Coulomb potential used to study and improve interpretability of Graph Attention Networks.

💡

Semiconductor ML & Photonics

High-performance ML for semiconductor defect analysis and hot-spot prediction (Siemens / Intel). Quantum Cascade Laser photonic device design and fabrication (Harvard / Northwestern).

Research Output

Selected Publications

2026 — NEW arXiv preprint · cs.AI

PLDR-LLMs Reason at Self-Organized Criticality

Shows that PLDR-LLMs achieve reasoning and generalization when long-range interactions overlap at criticality, leading to a global metastable steady state. Defines a global order parameter to quantify reasoning capability without curated benchmark datasets.

📄 arXiv:2603.23539 ⌕ Code (PyTorch) 🤗 Models

2025 arXiv preprint · cs.AI

PLDR-LLMs Learn A Generalizable Tensor Operator That Can Replace Its Own Deep Neural Net At Inference

Demonstrates that PLDR-LLMs learn a singularity condition as a rank-1 zero-determinant matrix, improving interpretability and enabling the deductive tensor to replace the deep neural net during inference.

📄 arXiv:2502.13502 ⌕ Code (PyTorch) 🤗 Models

2024 arXiv preprint · cs.AI / cs.LG

PLDR-LLM: Large Language Model from Power Law Decoder Representations

Introduces the PLDR-LLM decoder-only architecture using non-linear attention with learnable power law coefficients — a generalization of Scaled Dot-Product Attention with complete open-source HuggingFace implementation.

📄 arXiv:2410.16703 ⌕ Code (TensorFlow) 🤗 Models

2021 arXiv preprint · cs.LG

Power Law Graph Transformer for Machine Translation and Representation Learning

Demonstrates that non-linear attention with learnable power law coefficients can be trained as an encoder-decoder transformer for machine translation tasks.

📄 arXiv:2107.02039 ⌕ Code (TensorFlow/Keras)

2019 arXiv preprint · cs.LG

CoulGAT: An Experiment on Interpretability of Graph Attention Networks

Introduces attention inspired by screened Coulomb potential to interpret characteristics of Graph Attention Networks.

📄 arXiv:1912.08409 ⌕ Code (TensorFlow/Keras)

2005 – 2014 Applied Physics Letters · Optics Express · SPIE · Phys. Rev. Lett.

Quantum Cascade Laser & Photonics Publications (15+ papers)

Peer-reviewed journal papers on high-power photonic crystal distributed-feedback QCLs, tapered quantum cascade lasers, hyperspectral imaging, quantum entanglement, and photonic integration. Published in collaboration with Capasso Lab (Harvard), Razeghi Lab (Northwestern), MIT Lincoln Lab, and others. Total citations: 1035 · h-index: 14.

📄 Full Publication List (Google Scholar)

Career

Experience

Jan 2019 – Present

Fromthesky Research Labs LLC — Hillsboro, OR

AI/ML Researcher & Founder

Conducts fundamental AI research and develops practical industry solutions using large language models. Invented PLDR-LLM and PLGT architectures. Developed a complete open-source HuggingFace model ecosystem in PyTorch and TensorFlow for inference and training.

LLM Research PyTorch TensorFlow HuggingFace NLP Generative AI

Oct 2021 – Oct 2025

Siemens Digital Industries — Wilsonville, OR

Senior Machine Learning Engineer

Conducted pathfinding work in the Calibre group for root-cause analysis of microchip defects, optimization of semiconductor process parameters, and SEM/wafer-level defect analysis using advanced decision tree architectures, deep learning, and Shapley-value-based model interpretability.

Decision Trees Deep Learning Shapley Values scikit-learn Defect Analysis Python

Jun 2020 – Oct 2021
Jul 2012 – Dec 2016

Intel Corp — Hillsboro, OR

Process Integration Engineer

Managed backend process development for Intel's 10 nm (Cannonlake / Icelake / Tigerlake / Alderlake) and beyond technology nodes. Led nanoscale metal gapfill and interconnect (M0/M1+) process development; drove defect reduction using data-driven methods and cross-functional collaboration.

10 nm / 7 nm Process Integration SQL R / JMP DoE Defect Reduction

Dec 2016 – Jun 2020

Intel Corp — Hillsboro, OR

Physical Verification Software Engineer

Developed EDA runset code for interconnect DRC, latch-up reliability, and parasitic RC extraction for 7 nm and 10 nm nodes. Built regression models for unit testing of hierarchical EDA databases. Improved runtime and memory performance through benchmarking.

EDA ICV / PXL StarRC Calibre Python Perl / Tcl

Sep 2011 – Jul 2012

Harvard University — Cambridge, MA

Postdoctoral Fellow, Capasso Lab

Developed high-brightness Quantum Cascade Tapered Laser architecture for fast on-chip spectroscopy-at-a-distance applications. Designed, fabricated, and characterized photonic laser devices in a cleanroom environment using photonic simulation tools.

Quantum Cascade Laser Photonic Integration Cleanroom Spectroscopy

Sep 2004 – Jun 2011

Northwestern University — Evanston, IL

Research Assistant (PhD), Center for Quantum Devices

Developed record-breaking high-power photonic crystal distributed feedback QCL architectures. Achieved high-yield 3D photonic integration on III-V semiconductor wafers using holographic and e-beam lithography. Demonstrated on-chip single-mode DFB laser arrays for spectroscopy.

QCL / Photonic Crystal III-V Semiconductors E-beam Lithography Holographic Lithography

Academic Background

Education

PhD — 2011

Electrical Engineering & Computer Science

Northwestern University, Evanston, IL
Specialization: Quantum Cascade Laser Photonic Devices

BSc — 2003

Electrical & Electronics Engineering

Middle East Technical University, Ankara, Turkey

BSc Double Major — 2004

Computer Engineering