Research Accomplishment of the Month
Identifying
tunneling in ferromagneticinsulatorferromagnetic thin film structures
Johan J. Akerman, R. Escudero, D.
A. Rabson, C. Leighton, S. Kim, A. H. Romero,
Ivan K. Schuller, Renu Whig Dave,
J. M. Slaughter
Collaboration between the Nanoscience Group at UCSD and Motorola
Current interest in spindependent
electron tunneling is high due to promising applications such as magnetic
RAM and harddrive read heads.
The physical mechanism behind the large tunneling magnetoresistance (TMR) is the difference in electronic density of states at the Fermi energy in ferromagnetic materials. The tunneling probability is proportional to both the number of tunneling electrons and the number of final states. Since an electron with a given spin direction can only tunnel into an empty state with the same spin direction, a parallel magnetic state leads to high tunneling current (low R) and opposite magnetic states to low current (high MR). 
To increase the speed of a tunneling
device one reduces its resistance by using thinner and thinner tunneling
barriers.
With decreasing barrier thickness the probability of pinholes through the barrier increases where direct metallic conduction can short the tunneling current. Since pinholes exist on the nano scale they are extremely hard to detect microscopically over large sample areas. Instead one resorts to direct electrical measurements to probe the quality of the tunneling barrier. In the 1960s and 1970s a set of socalled Rowell criteria were formulated to determine whether tunneling dominates the conduction in superconducting tunneling devices. In the case of magnetic tunneling devices only three criteria remain: i) exponential thickness dependence

The first criterion is based on
the evanescent nature of the electron wave function inside of the insulating
barrier.
However, one can easily show that an entirely classical conduction mechanism via metallic shorts through barrier pinholes can mimic the tunneling behavior in the following way. Consider a perfectly smooth metallic surface on top of which we randomly deposit insulating blocks up to an average thickness m. It can be shown that the random deposition leads to a Poisson distribution of the local insulator height, and also to an uncovered metal area that is a fraction exp(m) of the total area. On top of this insulator we deposit another layer of metal and ‘switch on’ classical metallic conduction. The total current will be proportional to the uncovered area, hence exponentially dependent on the average insulatore thickness. As a consequence, the first criterion is not a reliable way of proving that tunneling dominates the conduction. 
To test the second criterion, we
made superconductorinsulatorferromagnet trilayer structures. The superconducting
layer (Nb) enables us to uniquely identify a good tunneling barrier since
tunneling into a superconductor below its critical temperature (Tc) will
reveal its very singular density of states.
Alternatively, direct conduction through a small point contact into a superconductor will lead to a completely opposite behavior, known as Andreev Reflection. We made hundreds of samples and divided them into good junctions and direct contacts using the difference below Tc. We then study the nonlinear currentvoltage relation of these samples at higher temperature and fit the data with the socalled Simmons model. We find that regardless of the presence of a short through the barrier, the fitted currentvoltage relation of a sample can give perfectly reasonable barrier parameter values. As a consequence, not even the second criterion should be used to prove that tunneling dominates the conduction. 
The third criterion is tested in
the same way. All samples are divided into good junctions or direct shorts
by measuring below Tc of the superconductor. The temperature dependence
of the resistance is subsequently determined.
We find a 100% correlation between a weak insulatinglike temperature dependence and a true tunnel junction. As a consequence, the third criterion is the ONLY criterion that is reliable and should hence ALWAYS be used to prove that tunneling indeed dominates the conduction in a given magneticinsulatormagnetic trilayer. 
To investigate whether additional
criteria may be formulated, we studies the temperature and bias dependence
of stateoftheart magnetic tunnel junctions both before and after the
insulating barrier had been shorted by applying a short voltage pulse above
its breakdown voltage.
The original junctions showed a weak insulatinglike temperature dependence of the resistance, which hence proves that they are indeed good junctions. After the voltage pulse, the same junction instead showed a metallike temperature dependence, clearly indicating that a metallic short is piercing the barrier. It is interesting to fit the currentvoltage relation both before and after the short was introduced. Whereas the original junction showed a very weak temperature dependence of the fitted barrier parameters, the short introduces an artificial temperature dependence to the apparent barrier parameters. The short effectively increases the apparent barrier thickness and decreases the apparent barrier height. This has important implications
in studies of ultrathin barriers. If one studies the barrier parameters
as a function of decreasing thickness and finds that the apparent thickness
is leveling off but the apparent height is rapidly decreasing – one then
knows that the barrier is too thin, i.e. a short is piercing through the
barrier.

Another interesting consequence
of the short is a dramatic increase in device noise and device instability.
Whereas noise in magnetic tunnel junctions is mostly of magnetic nature, the short will introduce additional Johnson noise since it is an ordinary metallic resistance. Moreover, the short will have to sustain a huge current density, up to orders of 1010 Acm2, and is likely to heat up locally, further increasing the Johnson noise. Finally, the device instability likely arises from electromigration due to the large current density. 
Read more:
'Reliability of normalstate currentvoltage
characteristics as an
'Pinholes may mimic tunneling'
'Tunneling criteria for magneticinsulatormagnetic
structures'
