About
I am a founding member of technical staff at Ricursive Intelligence, working on AI for chip design and the Ripple family of language models. Previously, I was a member of the AlphaChip team at Google Brain and DeepMind, contributing to foundational research on AI-driven chip design published in Nature. My work spans differentiable programming, physical design, and timing analysis for semiconductor chip design.
Broadly, I work at the intersection of machine learning and electronic design automation (EDA) — applying reinforcement learning, graph neural networks, and differentiable optimization to hard problems in chip placement and floorplanning. The goal is to make AI a practical, foundational part of how modern semiconductors are designed.
Experience
Develop AI for chip design and contribute to Ripple-65B Preview, focusing on reinforcement learning (RL) infrastructure, supervised fine-tuning (SFT), data generation, and model evaluation.
Member of Google DeepMind GenAI's AI for AI chip team. Advanced the state of the art in AI/ML for semiconductor design through differentiable and learning-based approaches to physical design and timing analysis.
Contributed to foundational research on ML-driven chip design as part of Google Brain's ML for Systems team. Designed a novel graph convolutional architecture enabling generalization across chip placement tasks — a core innovation of AlphaChip.
Designed and delivered scalable infrastructure for a two-tower deep learning retrieval system, improving ad latency and relevance.
Education
Supervisor: Prof. Farinaz Koushanfar (UC San Diego)
Selected Publications & Patents
- Mirhoseini, Goldie, Yazgan, Jiang, Songhori, E., et al. "A graph placement methodology for fast chip design." Nature 594 (2021): 207–212.
- Songhori, E., Wang, Mirhoseini, Goldie, et al. "Alignment Cost for Integrated Circuit Placement." U.S. Patent Application 17/890,370 (2024).
- Zhang, Safeen, Mirhoseini, Goldie, Songhori, E. "Full-stack hardware accelerator search." U.S. Patent Application 18/285,578 (2024).
- Goldie, Mirhoseini, Songhori, E., Jiang, Wang, et al. "Generating integrated circuit placements using neural networks." U.S. Patent Application 17/555,085 (2022).
- Riazi, Weinert, Tkachenko, Songhori, E., Schneider, Koushanfar. "Chameleon: A hybrid secure computation framework for ML applications." AsiaCCS (2018).
- Songhori, E., Hussain, Sadeghi, Schneider, Koushanfar. "TinyGarble: Highly compressed and scalable sequential garbled circuits." IEEE S&P (2015).
Awards & Honors
Projects
- Ripple-65B Preview (Hugging Face) — Ricursive Intelligence’s open-weight language model for reasoning and agentic coding.
- AlphaChip (Circuit Training) — open-source chip floorplanning framework using distributed deep reinforcement learning. I contributed to this repository.
- Mapping the Narrow Corridor with LLMs — using large language models to chart countries' paths through Acemoglu and Robinson's Narrow Corridor.