Apprehending Performance Variability in CNT Network Thin-Film-Transistors
Woodruff School Associate Professor Satish Kumar and ME graduate student Jilauo Chen have developed a method to capture the performance variability in current voltage characteristics of thin-film-transistors, a process that will be helpful for reliability analysis and testing of carbon nanotube (CNT)-thin-film-transistors (TFTs) based devices. This technique can be applied to improve low-cost gas sensors for monitoring pollution and controlling industrial emissions.
CNT network based TFTs are of high interest for low-cost and large area electronics such as antennae, RF tags, sensors, etc. Fabrication of these TFTs is compatible with both flexible and hard substrates, which makes them very promising for next generation of IoT and wearable devices.
Challenging reliability problems arise in device-to-device performance variability of these TFTs and is considered a bottleneck in the employment of these devices for various applications. This variability is rooted in the randomness of CNT networks, in the variation of individual CNT properties due to change in chirality, and fabrication imperfections. Kumar and Chen developed a novel method to capture the variability in current-voltage (I–V) characteristics of TFTs through a combination of experimental and theoretical analysis of the major sources that cause performance variation. This technique will enable re-construction of the performance variability of CNT-TFTs from the distribution functions of the relevant parameters, which will be very helpful for reliability analysis and testing of CNT-TFT based devices and circuits.
The research was supported by the NextFlex and National Science Foundation, and reported in the journal IEEE Transactions of Nanotechnology in the March 2018 issue.
CNTs are one dimensional material with exceptionally high carrier mobility, which makes them appealing for logic and computing using TFTs with < 10 nm characteristics length. In these applications, to enable high performance beyond Si based transistors, CNTs need to be aligned and semiconducting CNTs need to be separated out from metallic CNTs (typically 1/3rd of CNTs are metallic). These are significant technical challenges and make the fabrication process very expensive. On the other hand, for many large area electronic applications, such as sensors, antennae, RF tags, etc., the network of CNTs, a spaghetti like thin film, based TFTs is of high interest for low-cost fabrication on both hard and flexible substrates.
The operation of typical CNT network TFT is controlled by three electrodes: source, drain and gate. Source and drain electrodes make direct contact with the network of CNTs and gate electrode can be below or top of the CNT network separated by a thin dielectric. The implementation of thin gate dielectric with high dielectric constant is quite important as it leads to high gate capacitance and thereby increase drive current, switching speed and device performance, and also important for further miniaturization of FETs in its end-applications.
The fabrication of CNT-TFTs with channel length of 5-10 micrometers is a complex process; layer-by-layer fabrication is required and most of the steps are done in a cleanroom using photolithography and lift-off techniques. The fabrication starts with metallic gate electrode deposition, followed by deposition of TiO2 and HfO2 using atomic layer deposition (ALD), CNT network layer deposition using solution based techniques, source/drain electrode deposition and finally removal of CNTs by etching away unwanted part of the CNT network outside of TFT area. Between different layers of fabrication, surfaces treatment become very necessary, e.g., functionalization of surface by amine group for CNTs to better adhere and make uniform network on gate-dielectric to achieve good performance. We have used ALD technique to deposit 50-nm thick HfO2 as gate dielectric, and able to operate devices < 2V applied bias for high energy efficiency.
"In order to capture the variability of CNT-FET I-V characteristics, sets of FETs with different dimensions were fabricated and for each set 100s of FETs were fabricated. The next step was to identify the variables which have significant effect on I-V characteristics. We used an analytical model for I-V characteristics and identified the primary factors for variability are % of metallic-CNTs, threshold voltage at which devices turn on, CNT mean length, and CNT network density. A challenging problem was how to identify the variability in each of these parameters especially ones related with the CNT network," says Dr. Kumar. "For each variable used in the I-V relationship, the distribution function for variation are obtained statistically, e.g., to obtain variation of CNT mean length, a low density CNT network was grown, multiple SEM images are taken, all CNTs in these images are fitted with spline function to measure the lengths and resulted in a log-normal distribution. For CNT network density, the SEM images of network are converted to white-gray images to estimate coverage of surfaces by CNTs and a normal distribution of network density is observed. All these steps are quite tedious but crucial to capture the details of CNT network morphology and its variation."
In measured I-V characteristics of FETs, variability is a comprehensive effect caused by all the variables. We considered the comprehensive variations from all important sources of variation and established that more than 90% of the variation range in I-V characteristics, obtained through the analytical formulation of I-V, overlap with the experimental data for different channel length series. Basically, we are able to re-construct the variability in I-V of TFTs from the distribution functions of the relevant parameters. Large variability in device performance can overwhelm the design of circuits made using CNT-FETs and significantly challenge its reliability during operation. The work is quite beneficial in this context- first it will direct future research to control the processing conditions which can reduce the important sources of variability and second representing the variability in I-V of FETs through a distribution function of individual parameters in an analytical correlation will significantly ease the design and testing of circuits for different applications.
CNT-TFTs can be translated to flexible substrates or can be directly printed on flexible substrates. They can be used as low-cost gas sensors for monitoring different gases in low ppm and ppb level. The advantages of CNT-FETs based gas sensors mainly lie in their high sensitivity, fast response, and physical/chemical stability, which have enormous potential in pollution monitoring and industrial emission control. They also have high potential for artificial electronic skin, memory devices, stretchable electrode for actuators, etc. In all these cases, understanding and controlling variability for better performance reliability will be a key enabler.