研究発表

This work proposes an evaluation technique of self-heating (SH) in a 65-nm bulk CMOS transistor at cryogenic temperature (4.2K). For the in-situ monitoring of self-heating in cryo-CMOS circuits, transient response of the subthreshold drain current is used, which makes it possible to estimate the temperature at on-state of MOSFETs. The relation between the temperature increase by self-heating and power consumption of MOSFET has been experimentally obtained, which will help a design of high performance cryo-CMOS circuits for quantum computer application.

NanoBridge (NB) is a kind of electrochemical resistive-change device using Cu ion migration from Cu interconnects into electrolyte. NBs are integrated by only two additional masks in a standard Cu-BEOL without scarifying CMOS performance. The newly developed split Cu electrode enables the lower set voltage (2.2V) and superior retention, which have been demonstrated in a 28 nm-node. High density embedded nonvolatile NB memory macro in a 65nm technology is successfully developed with P E=30 ns 30 ns programming based on the highly reliable and low-cost manufacturable NB technology. High yield is achieved with satisfying 30ns read, and more than 10 years retention at 150°C. NB properties can be systematically adjusted by tailoring the NB stack, achieving to broaden NB applications.

A recent progress of NanoBridge (NB) technology is reviewed in terms of nonvolatile field-programmable gate array (FPGA) and memory (NVM) applications. The NB is a kind type of electrochemical resistive-change device using migration of Cu, in which a thickness of Cu bridge can be controlled by a programming current and an appropriate ON-conductance is programmed for depending on the applications. For a routing switch in the FPGA, two NBs are connected in series and each conductance is tuned to 0.8mS to transfer signals, in which the NBFPGA realizes 78% area and 80% power reductions compared with a conventional SRAM-FPGA. For the NVM application, that is tuned to 0.2mS, which has 150oC 20years retention with 30% degradation of the conductance, which is potentially applicable for automobile/AIoT applications.

An ON-state retention of a 40nm-node atom switch embedded nonvolatile memory (eNVM) has been carefully investigated for IoT/AI inference solution. Based on ON-conductance (Gon) tuning model of atom switch, one order of magnitude lower programming power is achieved while keeping the same Gon. Smaller ON-state retention dependences on temperature (Ea = 0.2eV) and time (n = 0.11) are experimentally clarified and the lifetime is predicted to be more than 10 years at 150°C under +20% shift criteria of Gon.