1: \begin{abstract}

2:   Spin-transfer torque magnetoresistive random access memory is a

3:   potentially transformative technology in the non-volatile memory

4:   market. Its viability depends, in part, on one's ability to

5:   predictably induce or prevent switching; however, thermal

6:   fluctuations cause small but important errors in both the writing

7:   and reading processes.  Computing these very small probabilities for

8:   magnetic nanodevices using naive Monte Carlo simulations is

9:   essentially impossible due to their slow statistical convergence,

10:   but variance reduction techniques can offer an effective way to

11:   improve their efficiency.  Here, we provide an illustration of

12:     how importance sampling can be efficiently used to estimate

13:   low read and write soft error rates of macrospin and

14:   coupled-spin systems.

15: \end{abstract}

16: