RRAM is generally considered as the next generation of nonvolatile memory due to its simple structure, low power consumption, high-speed, high density and its compatibility with conventional CMOS processing. Another attractive feature of RRAM is its tunable resistance with applied voltage, making it a promising candidate for multi-level cell (MLC) storage and for neuromorphic computing. However, reliability and variability issues limit the potential of RRAM. Particular for MLC and neuromorphic application, poor uniformity poses as a severe hindrance. This variability is related to the stochastic nature of, filament formation, resistance switching and poor control of switching energies. While material engineering and structural optimization techniques are employed to reduce the intrinsic randomness in RRAM, these strategies are ineffective when programming energy is not under control. In this work, we propose a Compliance-free Ultra-Short Smart Pulse Programming (CUSPP) technique which offers improved control of FORMING/SET/RESET energies. This technique can potentially lead to tighter distribution of critical RRAM parameters, such as RON and ROFF, resulting in well-defined resistance window during extended cycling. CUSPP is a custom-built fast pulse-train programming and measurement system with real-time resistance sensing and automatic cycling between SET and RESET when specific resistance is reached. An ultra-short (~100 ps) pulse generator with controllable repetition rate and pulse height (up to 6V) is our source. The custom-built circuit continuously monitors the device resistance using a low DC sensing voltage to determine if the SET/RESET resistance is achieved, and to automatically change the pulse train’s height and/or polarity to enable rapidly cycling up to 2.5 million cycles per second. Unique to CUSPP measurement setup is the absence of a current limiting element such as a series resistor or transistor. The ultra-short pulse width limits the duration of current flow through the device and provide a tight control of switching energies. In the current version of the CUSPP, resistance programming is solely achieved by repetitive pulsing the device with constant pulse amplitude and pulse width. This is in contrast to many reported programming protocols where pulse amplitude and/or width are incrementally increased to reach the target resistance state. In targeting RON/ROFF, higher precision can be achieved by lowering the pulse amplitude and a concomitant higher number of pulses. Thus, high precision comes at a cost of programming speed. With the pulse repetition rate up to a few MHz this cost is entirely acceptable for mechanism studies. CUSPP is a versatile tool that can perform all essential programming of the RRAM with precision and speed while allowing the experimenter to look at the filament kinetics in details.
Read full abstract