The Institute of Microelectronics at Peking University committed itself to develop advanced devices and new materials for silicon micro/nanoelectronics, which considerably supports the development of the integrated circuit industry in China in the past ten years. The main contributions are included as follows.

1) Ultra-low-power devices.

Research group in the Institute of Microelectronics have dedicated to the research of advanced low-power devices and related key technologies in the past few years. From the perspective of structure engineering, we have proposed a new body-on-insulator (BOI) FinFET structure based on bulk Si substrate, effectively suppressing the source/drain leakage beneath the Fin channel and experimentally obtaining large ON/OFF ratio of more than 10⁸. Moreover, we have successfully fabricated gate-all-around silicon nanowire transistors (SNWTs) with 10nm diameter on bulk substrate by fully epi-free CMOS-compatible technology, which demonstrate the best performance among SNWTs at that time. SNWT-based analog circuit of current mirror was also first experimentally demonstrated, showing its unique potential at subthreshold operation for low-power applications. From the perspective of mechanism engineering, tWe first proposed a new kind of hybrid operation mechanism of Schottky injection and band-to-band tunneling, and have proposed novel T-gate/multi-finger-gate Schottky barrier tunnel FET (TSB-TFET/MFSB-TFET)(6 IEDM/VLSI papers). The proposed device can achieve high ON-current from the dominant Schottky current, effectively reduced OFF-current from the introduced self-depletion effect, and steep subthreshold swing from dominant band-to-band tunneling current with enhanced electric field simultaneously. This hybrid operation mechanism fundamentally addresses the issue of low drive current in conventional TFETs, and obtains higher ON-current and lower subthreshold swing while maintaining ultra-low OFF-current. Based on the CMOS-compatible process, the fabricated MFSB-TFET exhibits superior performance with subthreshold swing of 29mV/decade, which is the lowest value in silicon-based tunnel transistors ever obtained on record. The ON-current is also improved by 2.5 decades compared with conventional TFET, and large ON-OFF current ratio of more than 10⁸ is achieved. In addition, the MFSB-TFET also demonstrates comprehensive electric properties enhancement in terms of output behavior, capacitance, delay, gain, noise, variability and reliability, indicating its great potential for ultra-low-power applications. The novel tunneling technology has been transferred to Semiconductor Manufacturing International Corporation and is being developed on standard foundry platform, which will provide an important foundation for the development of ultra-low power integrated circuit technology.

2) Resistive random access memory.

Resistive random access memory (RRAM) is one of the most promising candidates for next generation nonvolatile memory due to its superior performance and scalability. We explored the device structure, materials, model and simulation of various RRAM devices, including metal-oxide based RRAM devices and flexible parylene-based organic RRAM device. We proposed a unified microscopic physical principle based on oxygen vacancy filament conducting model to clarify resistive switching behaviors of metal oxide based RRAM and series of technical methods to improve and optimize the resistive switching uniformity and reliability of metal oxide based RRAM, including material doping, interfacial layer design, and operation mode innovation. A RRAM SPICE model was also developed to capture direct current and transient behavior of metal-oxide based RRAM,including descriptions of parasitic effect, thermal effect and intrinsic variation effects. The RRAM SPICE model has been applied to design and optimize the RRAM array, nonvolatile logic and neuromorphic computing circuits. Monte Carlo method was proposed to simulate the switch behavior of OxRRAM and CBRAM. The evolution of microcosmic features during the resistive switching and the reliability degradation process can be reproduced along with the corresponding electrical characteristics. Method to simulate the charge trapping memory from cell to array was proposed, which can help to understand the CTM characteristics such as the charge loss, retention, disturb and their impact on the array performance. 1Mbit~64Mbit RRAM test chips were also successfully demonstrated in SSD application.

3) Variability and Reliability.

When CMOS technology scaling into nanometer region, device variability and reliability (or, aging) are becoming crucial for circuit robustness and design-technology co-optimization. PKU has contributed deeply in the characterization and theoretical modeling of line-edge/width roughness (LER/LWR). For example, PKU was the first to point out the cross-correlation of LER. Based on this, PKU further proposed the first full-set compact model for static random variations in FinFET, including Fin-edge roughness, gate-edge roughness, as well as metal-gate granularity, which can be embedded into industry-standard BSIM-CMG model, thus is very helpful for FinFET circuit design. PKU was among the first to point out the stochastic NBTI effect in transistor aging. Thus, for the aging-induced dynamic variation, apart from conventional time-dependent device-to-device variation, PKU found that this effect induces a new source of cycle-to-cycle variation due to the random occupation of trap states in each operation cycle during circuit aging based on the related results; more than 15 IEDM/VLSI papers were published. The related achievements have been transferred to SMIC, Cadence and Hisilicon to promote the technology R&D and circuit design methodology for advanced technology nodes.

4) Neuromorphic devices.

Brain-inspired neuromorphic computing is expected to revolutionize the architecture of conventional digital computers and lead to new generation of intelligent, power efficient computing paradigm, where neuromorphic devices serve as the building blocks. Research group in the Institute of Microelectronics have performed extensive studies on the design and fabrication of novel neuromorphic devices as well as their integration and network construction toward neuromorphic applications. A novel 3-terminal memristive device was proposed with planar and vertical configurations that could implement heterosynaptic plasticity, an important learning rule in biological systems, so that they can be used as the building blocks of neuromorphic systems. In addition, it has been experimentally demonstrated physically evolving networks in nanoscale, solid-state, multi-terminal memristive devices, and developed systematic approaches for improving the analog switching linearity in memristive synapses. A fuzzy restricted Boltzmann machine network was also proposed that could largely tolerate device variations in neuromorphic elements. Large-scale integration of memristive synapses faces great challenges due to the sneak path problem. Novel memristors were developed that can suppress the sneak currents by adopting a novel oxide heterostructure where each oxide layer can be tailored for a specific function during resistance switching, and by employing functionalized graphene as the electrode that has threshold switching characteristics and serve as internal selector element. These studies have been published in high-profile journals, and have been well received and cited by the research community.

Related works have been published in Nature Comm., Adv. Mater., ACS Nano, Nanoscale, IEEE EDL and IEEE T-ED, IEDM, VLSI and other top tier journals and conferences in the microelectronics fields. In the past 10 years, more than 40 papers have been published in IEDM, which is more than 50% of the publications from the mainland China. Several works have been involved in the International Technology Roadmap for Semiconductors (ITRS 2009, 2011, 2013, 2015). The research achievements won the second prize of National Technology Invention Award (2010) and the first prize of Natural Science Award, the Ministry of Education (2015). Some of the above results have been transferred to SMIC, Hisilion, Cadence and Synopsys.