Imec in Belgium has developed the first modelling framework to accurately predict 3D thermal transport in RF devices for 5G and 6G networks.
The research, presented at the 2022 International Electron Devices Meeting this week, uses a Monte Carlo Boltzmann framework to model microscopic heat carrier distributions and so predict 3D thermal transport in the devices.
Case studies with GaN high-electron-mobility transistors (HEMTs) and InP heterojunction bipolar transistors (HBTs) revealed peak temperature rises that are up to three times larger than conventional predictions with bulk material properties. The new tool will be key to improving the thermal performane of more powerful RF devices.
To optimize GaN and InP devices for RF applications and make them cost-effective, much attention is paid to upscaling the materials for a silicon platform and making them CMOS compatible. However, with shrinking feature sizes and rising power levels, self-heating has become a major reliability concern, potentially limiting further RF device scaling.
“Tuning the design of GaN- and InP-based devices for optimal electrical performance often worsens thermal performance at high operating frequencies,” said Nadine Collaert, programme director of advanced RF at imec. “For GaN-on-Si devices, for example, we recently achieved tremendous progress in electrical performance, bringing the power-added efficiencies and output power for the first time on par with that of GaN-on-silicon carbide (SiC).”
“Further enlarging device operating frequency will require downsizing the existing architectures. In these confined multilayer structures, however, thermal transport is no longer diffusive, challenging accurate self-heating predictions,” she said.
“Our novel simulation framework, yielding good matches with our GaN-on-Si thermal measurements, revealed peak temperature rises up to three times larger than previously predicted. It will provide guidance in optimizing these RF device layouts early in the development phase to ensure the right trade-off between electrical and thermal performance.”
Such guidance also proves very valuable for the novel InP BTs, where imec’s modeling framework highlights the substantial impact non-diffusive transport has on self-heating in complex scaled architectures.
For these devices, nanoridge engineering (NRE) is an interesting heterogeneous integration approach from an electrical performance point of view.
“While the tapered ridge bottoms enable low defect density within the III-V materials, they, however, induce a thermal bottleneck for heat removal towards the substrate,” said Bjorn Vermeersch, principal member of technical staff in the thermal modeling and characterization team at imec.
“Our 3D Monte Carlo simulations of NRE InP HBTs indicate that the ridge topology raises the thermal resistance by over 20 percent compared to a hypothetical monolithic mesa of the same height. Our analyses furthermore highlight the direct impact of the ridge material (e.g., InP vs. InGaAs) on self-heating, providing an additional knob to improve the designs thermally.”
More IEDM 2022 articles:
- TSMC heads below 1nm with 2D transistors at IEDM
- IEDM: TSMC to report 2D nanosheet transistor
- Intel looks to 1nm for trillion transistor ‘chips’