Advanced Metrology for Characterization of Magnetic Tunnel Junctions

For the past 50 years the prediction that turned into what is now known as Moore’s Law has held true. It states that the complexity of integrated circuits such as computer processors will double approximately every second year. In order to keep up the industry has been on constant lookout for new solutions to take products to the next node and the field of memory technology is no exception. Over the past decade research and development in a novel, non-volatile memory type known as MRAM has intensified, and commercial MRAM devices are now available. MRAM holds an extremely favorable position as it is believed to have the potential of becoming a truly universal memory solution dominant within all fields of memory application. A decade ago the company CAPRES A/S introduced the so-called CIPTech, which is a metrology tool utilizing micro four-point probes (M4PPs) and a method known as current in-plane tunneling (CIPT) for characterization of magnetic tunnel junctions (MTJs), which constitutes the key component not only in MRAM but also the read-heads of modern hard disk drives. MTJs are described by their tunnel magnetoresistance (TMR), which is the relative difference of the resistance area products (RA) at two characteristic resistance levels (high and low) of the MTJ device. In the final memory application these resistance states correspond to a digital “1” or “0” stored. During CIPT measurements the tool will alter the state of the MTJ by application of an external magnetic field. With the CIPTech the turn-around time for measurements on magnetic tunnel junctions shortened dramatically from two days to one or two minutes. As one happy user put it, it was like going from a tricycle to a Ferrari in one step, and the tool is now in use in all major memory companies throughout the world. However, with a measurement time of 1-2 minutes per measurement, the technique is commonly used just for research and development of novel MTJ stacks and not for full wafer analysis, which would otherwise provide valuable information with respect to uniformity, e.g. for tool optimization. The precision of CIPT measurements is limited by electrode position errors, the importance of which increases for decreasing electrode pitch. This is a challenge to the measurement method as such and may become even more so in the future, when the cell size of MRAM is scaled down to increase memory density. The fundamental goal of this project has been to provide cheaper, faster and more precise metrology for MTJs.
This goal has been achieved in part by the demonstration of a static field CIPT method, which allows us to reduce the measurement time by a factor of 5, by measuring only RA thus excluding TMR. This enhancement is obtained purely by acquiring only half of the data needed for the conventional switching field CIPT measurement and particularly by avoiding magnetic field switching. We observe that the new method measures essentially the same RA values as compared to the conventional strategy. By offering the choice of characterizing either RA_low or RA_high the static field CIPT method has an added advantage over the conventional switching field CIPT method, which relies on the characterization of both RA values. This allows for an improved matching of the range of available electrode pitches and sample transfer lengths, which may effectively increases the dynamic range of any given micro 12-point probe (M12PP).
Without the requirement for switching magnetic fields during measurements the static field CIPT method has inspired the concept of detached magnet setups for future CIPTech tools. While lowering the complexity of the measurement system a detached magnet setup, e.g. a proposed letterbox magnet, could provide superior dynamic range and field homogeneity as compared to current state of the art solutions. We have carried out an extensive characterization of electrode position errors and experimentally shown that the dominant sources of error in single configuration micro four-point probe resistance measurements are in-line probe geometry errors and in-line static position errors. These errors were shown to be eliminated very effectively using dual-configuration measurements and position error correction algorithms. The standard deviation of the static in-line position error for measurements with Au coated electrodes on Ru thin film samples was found to be in the range from 3.9 nm to 7.5 nm. The standard deviation of the dynamic in-line position error was shown to be small ~3 Å and only detectable in measurements with high measurement current. At lower measurement currents the electrical measurement noise was the dominant error source. No significant ageing effect on position errors (except for a very slight reduction in position error with measurement age) was observed for a probe in the course of 5000 measurements. We have demonstrated how new probe designs may be evaluated and benchmarked against each other using the same strategy. Based on Monte Carlo simulations we have studied the influence of electrical noise as well as static and dynamic, in-line and off-line electrode position errors on four-point resistance measurements on MTJs. This study points out the van der Pauw position correction strategy based on combined measurement in four-point configuration A and C as being the most effective method to lower the relative standard deviation on the measured resistance. In line with this we find that the same method also provides the broadest dynamic range for the M12PP used in this project. As a means to further enhance the measurement precision we have proposed the addition of more subprobes of nominally identical electrode spacing and shown, that for one added sub-probe, the option for which two sub-probes shares two pins, yields the most significant reduction of electrode positional errors. Finally, a radical probe design entirely occupied by equidistant electrodes was proposed.