Understanding of MRAM-probabilistic-bit dynamics

We are in the revolutionary era of artificial/machine intelligence where day to day complex tasks is getting solved by intelligent machine networks. These networks incur a huge amount of computation cost for model training and execution which is increasing exponentially and becoming unsustainable. These traditional machine intelligent networks operate on deterministic way therefore requires huge data set and many steps for training to provide the confidence about the results and to avoid the false positive response, which results in an exponential increase in the computation. Contrarily, if network can be trained by incorporating uncertainty or stochasticity then requirements for giant data set and training steps can be reduced significantly.
 
The uncertainty or stochasticity in the machine intelligence networks can be implemented by random number generator (RNG). Tunable RNG is called the ‘probabilistic bit (p-bit)’ which is a key element to add the uncertainty in machine intelligence network. Traditional CMOS RNG, e.g., LFSR, are not energy and area efficient which limits the machine intelligence network scalability. In contrast, magnetic random-access memory (MRAM) device with low energy barrier can provide random number samples, which can be tune by bias to device third terminal to create probabilistic bit. The MRAM p-bit promises to be order more energy efficient and compact; hence allows to conceptualize a scalable system exploiting reconfigurable random number samples distribution. The outputs of the MRAM p-bit devices can be smartly combined in machine intelligent network to make them energy/throughput efficient. 
 
In this thesis work, the student will study the MRAM devices for the random number generations. Student will perform the detailed electrical characterization of the MRAM devices for switching dynamics, autocorrelations, devices variabilities. It is also expected from the student to perform the micromagnetic modelling on the devices for better understanding of memory device switching mechanism. Student may also extent study to the level of device SPICE simulation. 
 
This master thesis/internship work on MRAM device will involve the following task:

1. Electrical testing of the MRAM devices for random number generation
2. Micro magnetic modelling of the MRAM device dynamics 
3. MRAM based generated random numbers autocorrelation and NIST RNG test. Its benchmarking with state of arts implementations.

This thesis work allows the student to learn the dynamics of MRAM device for random number generation. Student will gain the in-depth understanding of the MRAM device physics, and its detailed characterizations. 

Type of work: 10% literature study, 20% modelling, 70% experiments

Type of Project:  Combination of internship and thesis 

Master's degree: Master of Engineering Technology; Master of Science 

Master program:    Nanoscience & Nanotechnology; Physics; Electrotechnics/Electrical Engineering 

Duration: 9 months  

Supervisor: Marian Verhelst (EE) 

For more information or application, please contact Ankit Kumar (ankit.kumar@imec.be).

Imec allowance will be provided for students studying at a non-Belgian university.