Search for: All records

Contact engineering on monolayer layer (ML) semiconducting transition metal dichalcogenides (TMDs) is considered the most challenging problem towards using these materials as a transistor channel in future advanced technology nodes. The typically observed strong Femi level pinning induced in part by the reaction of the source/drain contact metal and the ML TMD frequently results in a large Schottky barrier height, which limits the electrical performance of ML TMD field-effect transistors (FETs). However, at a microscopic level, little is known about how interface defects or reaction sites impact the electrical performance of ML TMD FETs. In this work, we have performed statistically meaningful electrical measurements on at least 120 FETs combined with careful surface analysis to unveil contact resistance dependencies on the interface chemistry. In particular, we achieved a low contact resistance for ML MoS2 FETs with ultra-high vacuum (UHV, 3×10-11 mbar) deposited Ni contacts, ~500 ohm·μm, which is 5 times lower than the contact resistance achieved when deposited at high vacuum (HV, 3×10-6 mbar) conditions. These electrical results strongly correlate with our surface analysis observations. X-ray photoelectron spectroscopy (XPS) revealed significant bonding species between Ni and MoS2 under UHV conditions compared to HV. We also studied the Bi/MoS2 interface under UHV and HV deposition conditions. Different from the case of Ni, we do not observe a difference in contact resistance or interface chemistry between contacts deposited under UHV and HV. Finally, this article also explores the thermal stability and reliability of the two contact metals employed here. 

models, it is difficult to fit and train a complete copy of the model on a single computational device with limited capability. Therefore, large neural networks are usually trained on a mixture of devices, including multiple CPUs and GPUs, of which the computational speed and efficiency are drastically affected by how these models are partitioned and placed on the devices. In this paper, we propose Mars, a novel design to find efficient placements for large models. Mars leverages a self-supervised graph neural network pre-training framework to generate node representations for operations, which is able to capture the topological properties of the computational graph. Then, a sequence-to-sequence neural network is applied to split large models into small segments so that Mars can predict the placements sequentially. Novel optimizations have been applied in the placer design to achieve the best possible performance in terms of the time needed to complete training the agent for placing models with very large sizes. We deployed and evaluated Mars on benchmarks involving Inception-V3, GNMT, and BERT models. Extensive experimental results show that Mars can achieve up to 27.2% and 2.7% speedup of per-step training time than the state-of-the-art for GNMT and BERT models, respectively. We also show that with self-supervised graph neural network pretraining, our design achieves the fastest speed in discovering the optimal placement for Inception-V3.