Discovered Materials是一家获Y Combinator孵化的初创公司，专门利用AI agents为半导体行业发现新材料[1]。该公司通过测试Anthropic、OpenAI和Kimi的7个模型，成功运用AI计算发现了动力学稳定且性能前景良好的新材料[1]。据介绍，AI在8小时内完成的材料发现工作相当于博士研究生数周的研究成果[1]。
公司重点针对芯片行业的关键痛点开展研发[1]。随着芯片功耗不断升高——英伟达H100芯片TDP为700W（2022年），Blackwell达到1.2 kW（2024年），Rubin预计将达2.3 kW（2026年）——GPU散热已成紧迫问题[1]。此外，3D芯片封装中的介电材料同样存在挑战，而采用3D封装技术可将数据移动能耗降低10-50倍[1]。公司已发布数百种新材料发现成果和性能基准测试[1]。据报道，该公司在过去3个月内模拟、合成并测试的热界面材料性能与全球最大化工公司保留20年以上的商业机密相当[1]。
公司创始人包括首席执行官Akash，拥有斯坦福材料科学博士学位且在半导体材料研究领域积累了11年经验[1]，以及联合创始人Advaith，在Carnegie Mellon学习AI，曾在Persona AI和Luma Labs任职[1]。
Discovered Materials, a startup incubated by Y Combinator, is leveraging artificial intelligence agents to accelerate the discovery of novel materials for the semiconductor industry [1]. The company has successfully identified kinetically stable materials with promising performance characteristics by testing seven models from Anthropic, OpenAI, and Kimi, achieving results equivalent to weeks of work by a PhD researcher in a matter of hours [1].
The startup has published hundreds of new material discoveries and corresponding performance benchmarks, with a particular focus on addressing two critical challenges in modern chip manufacturing: thermal dissipation for graphics processing units and dielectric materials for three-dimensional chip packaging [1]. The urgency of these challenges is underscored by rapidly escalating power demands—H100 processors consumed 700 watts of thermal design power in 2022, while Blackwell chips reached 1.2 kilowatts in 2024, and Rubin processors are projected to require 2.3 kilowatts by 2026 [1]. Three-dimensional packaging approaches offer potential energy savings of 10 to 50 times reduction in data movement power consumption [1].
The founding team brings substantial expertise to the venture: Akash holds a PhD in materials science from Stanford and brings eleven years of research experience in semiconductor materials, while Advaith studied artificial intelligence at Carnegie Mellon and previously worked at Persona AI and Luma Labs [1]. In the past three months alone, the team has simulated, synthesized, and tested thermal interface materials whose performance matches proprietary formulations that major global chemical companies have protected as trade secrets for over two decades [1].