| 研究実績の概要 |
In 2019, I collaborated with a research laboratory at Purdue University (USA) to achieve an experimental demonstration of probabilistic computing enabled by stochastic magnetic tunnel junctions (MTJ). The system was able to show integer factorization for integers up to 945 using 8 probabilistic bits, or p-bits. Other schemes such as quantum computing can produce similar results, but struggle with requirements such as operation at low temperature and control of quantum effects on large scale systems. The research began by investigating stochastic MTJs, which were engineered from a market standard STT-MRAM stack used in non-volatile memory applications. Unlike STT-MRAM, which has a large energy barrier separating two stable resistance states, I engineered the materials to show thermal fluctuations on the millisecond scale. The stochastic MTJ was then connected with circuit components to form the p-bit which operate as a tunable random number generator. After modifying a quantum computing algorithm, 8 p-bits form a Boltzmann machine where an input to one p-bit is a function of the fluctuations of all other p-bits. The interconnected system represents the lowest energies of the system as the highest probabilities, in this case, the answer to our integer factorization problem. The results from the spintronics-enabled probabilistic computing experiment open up an easily-accessible route to solving computationally complex problems, previously believed to require significant amount of time and resources to solve.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
In my intital application, I stated that in my tenure I would demonstrate integer factorization using stochastic magnetic tunnel junctions as probabilistic bits or bits. Because those results were completed, but because further development and a full investigation into the potential of this scheme has yet to be discerned, I give the status a (2).
|
| 今後の研究の推進方策 |
The results in 2019 set the groundwork for an expansive application space to explore. One of the first unknowns to investigate is how well does the probabilistic computing system using spintronics technology scale in area, power, and computation time. Concretely, the next goal is to determine the capabilities of the system at sizes suited for practical applications. Fabrication of such large scale circuits are out of the scope of my laboratory, so simulations that closely resemble the experimental results will be performed. I will look closely at what parameters affect the success of the system and essentially what guidelines need to be set when designing a spintronics-based probabilistic computer.
|
