材料科学
光电子学
热的
晶体管
热阻
宽禁带半导体
氮化镓
电子工程
场效应晶体管
硅
温度测量
逻辑门
集成电路
电子设备和系统的热管理
砷化镓
热导率
电路设计
电气工程
作者
Seokjun Kim,Daniel C. Shoemaker,Husam M. Walwil,Kyuhwe Kang,Jeong-kyu Kim,Kelly Woo,Bill Zivasatienraj,Puneet Srivastava,Isaac H. Wildeson,Srabanti Chowdhury,Sukwon Choi
标识
DOI:10.1109/ted.2026.3667665
摘要
In this work, the junction-to-heat sink thermal resistance (R${}_{\text {th}}\text {)}$of single-finger GaN high-electron-mobility transistors (HEMTs) mounted on a high-power laminate CuMo base package was analyzed using micro-Raman thermometry and electrothermal modeling. Analysis of the internal temperature distribution reveals that the die-attach material, substrate, and the semi-insulating buffer layer, respectively, comprise 2%, 28%, and 63% of the total Rthof a packaged device. To reduce the Rthof the device, the use of a high$\boldsymbol {\kappa }$(22 W/m$\boldsymbol {\cdot } $K) silver epoxy or AuSn solder (57 W/m$\boldsymbol {\cdot }$K) as the die-attach material is recommended. Simulation results indicate that replacing the substrate with single-crystal diamond lowers Rthby 21%, compared to a standard GaN-on-SiC HEMT. Regarding buffer engineering, a thinner semi-insulating GaN buffer layer leads to a reduced Rthunder a fully open channel bias condition. In contrast, under a partially pinched-off bias condition, reducing the thickness of the buffer layer from$1.8~\boldsymbol {\mu }$m to 400 nm leads to a ~10% increase in peak channel temperature rise. This finding was confirmed by performing nanoparticle-assisted Raman thermometry on transfer length method (TLM) structures with varying channel lengths, which demonstrated that buffer layer thickness optimization must account for the heat flux distribution and device geometry. Simulation results show that deposition of a polycrystalline (PC) diamond top-side heat spreader can lower Rthof a GaN-on-SiC HEMT by 17%. The findings of this work provide insight into the thermal design optimization of GaN monolithic microwave integrated circuits (MMICs).
科研通智能强力驱动
Strongly Powered by AbleSci AI