News | Group X | Jiang Xiao

A Fast Random Number Generator based on MTJ

Sat, 27 Aug 2022 08:15:12 +0800

Random numbers are essential in many information technologies, especially those concerning cyber security. Nowadays, most random numbers are the pseudo random numbers generated by mathematical algorithms. A compact, fast, and true random number generator (TRNG) is highly desirable.

Left: A dice is a random nubmer generator. Right: A MTJ memory element becomes a TRNG when the fluctaution of the free layer magnetization becomes fast.

It has been proposed that a TRNG can be constructed from a magnetic tunnel junction (MTJ), utilizing the thermal activated random magnetization switching. However, the MTJ-based TRNG can only operate at a relatively low frequency of 1KHz. Many efforts have been tried to enhance its generation rate.

In our recent paper [1], we demonstrate theoretically and confirmed by simulation that an application of a millitesla magnetic field can significantly reduce the potential barrier between the two equilibrium states, therefore enhance the random number generation rate of an MTJ-based TRNG by many orders of magnitude, from KHz to GHz. More importantly, the applied field has little effect on the read-out of the random numbers in the MTJ.

张家斌、陈锡恒硕士论文答辩

Mon, 23 May 2022 09:51:12 +0800

2022年5月23日，张家斌和陈锡恒两位同学成功举行了线上硕士论文答辩，论文题目分别是《基于COMSOL的时域与频域微磁学模拟研究》和《基于微磁学理论的磁结构调控与应用》。答辩委员会成员包括同济刘要稳教授、苏州纳米所曾中明研究院、天津大学兰金副教授、复旦大学吴义政教授。

两位同学经过共两个多小时的报告和质询答辩，答辩委员会一致建议授予张家斌和陈锡恒两位同学硕士学位。恭喜张家斌和陈锡恒！

A Hopfield neural network in magnetic films with natural learning

Fri, 19 Nov 2021 08:15:12 +0800

In order to develop its ultimate potential, artificial intelligence must shift from the software-based towards hardware-based neuromorphic algorithms. For the artificial neural network, it requires materials and devices that are capable of emulating neuron-like activation and synapses with reprogrammable weights. To the best of our knowledge, all existing approaches in implementing synapses require external computation to train the network. It would be highly desired that the hardware can naturally learn by itself with less or even without external intervention.

We distinguish two stages for natural learning based on physical systems: i) the physical systems CAN be adjusted via external stimulus; ii) the physical systems KNOW HOW to adjust. Most physical systems, including the memristor, can realize stage i), i.e. the stored weights in memristor can be modified via external stimulus. However, most physical systems don’t know how to adjust, so they need algorithms (such as the back-propagation algorithm) carried out by external computers to calculate the amount of the adjustment.

An adiabatic method to train Binary Neural Networks

Thu, 07 Oct 2021 08:15:12 +0800

We proposed a new training method to realize binary neural networks. [1] This method can successfully train a conventional neural network into a binary one, where the neuron activations and/or weights take binary values of (+1 and -1).

This work turned out to be compeletely in the field of Computer Science. It has little to do with Physics except the adiabatic method that is often used in physics studies. In fact, our initial goal was to try realize artificial neural network in a physical system based on MTJ-based MRAMs. But, in order to simplify the hardware, it is much more efficient to binarize the network first.

Compare with the conventional neural network using floating-point weights and neuron activations, the binary neural network uses binary values (0 or 1) for its weights and activations. Not only can the binarized network save much of the storage, it also speeds up the inference significantly because the heavy floating point matrix multiplication can be replaced by simple bitwise XNOR operations.

Skew scattering and side jump of spin wave across magnetic texture

Tue, 16 Mar 2021 08:15:12 +0800

Continuing our investigations on the interplay between spin wave and magnetic texture, we proceed to simplify their interplay by treating both the magnetic texture and spin wave as particle-like objects, and their interaction becomes a particle-colliding process.

In magnetic system, spin wave and magnetic texture can be regarded as two types of elementary excitations with fast and slow dynamics upon homogenous background. Understanding their interaction lies in the heart of our pursuit of constructing purely magnetic logic devices.

A powerful tool to describe the dynamics of the soft objects is the collective coordinate method, which focuses on the primary dynamics and ignores the minor deformations. In this work, we use the collective coordinates to describe the magnetic texture and the spin wave packet simultaneously. Based on this new approach, we identify two basic scattering patterns of spin wave across the magnetic texture: the magnonic skew scattering and the side jump, which are quite similar to the electron scattering with impurities in metals. As shown in the figure on the right, spin wave packet expriences the skew scattering as scattered by a magnetic skyrmion - a topologically non-trivial magnetic texture, while the side jump as scattered by a magnetic bubble - a topologically trivial texture. This extension of skew scattering and side jump concepts from electronic to magnetic systems allow us to obtain intuitive understandings on magnetic phenomena.

陈恭正硕士论文答辩

Sat, 25 Jul 2020 09:51:12 +0800

2020年7月25日，陈恭正同学成功举行了他的硕士论文答辩，论文题目是《铁电畴壁动力学的数值模拟研究》。陈恭正的答辩委员会成员包括同济刘要稳教授，成都电子科大严鹏教授，以及天津大学兰金教授。

经过一个多小时的报告和质询答辩，答辩委员会一致建议授予陈恭正同学硕士学位。恭喜陈恭正！

Loop theory for scattering in cavities

Sun, 19 Apr 2020 08:15:12 +0800

Continuing our pursuit in understanding coupled systems, we developed a simple graphical loop theory in calculating transmission and reflection for a general coupled systems with multiple internal modes. This graphical method is so simple that no calculation is needed.

For its important role in the fundamental physics as well as its intriguing docking with the quantum information technology, understanding and manipulating the light-matter interaction ranging from macroscopic scale to nanoscale is a long-lasting topic in many disciplines of physics, including the condensed matter physics, atomic, molecular and optical physics. A common knob to explore the interaction of light and matter is to put the matter into a high-quality cavity and measure the transmission/reflection spectrum.

Building an input-output formalism is critical to interpret the underlying physics of such hybrid systems. The existing theory usually handles the cavity systems with only one internal mode. However, the recent rising physics in the cavity-matter system naturally involves the contribution of multi-modes from the cavity and matter (see figure on the right), and the present theory becomes unmanageably complex as the number of modes increases.

Spin wave computing

Sat, 22 Feb 2020 20:15:12 +0800

The two main issues of the present computing technology are the increasingly severe energy dissipation and the limited computing speed. The origin of both issues can be traced back to the von-Neumann computing architecture, in which the data storage and processing are separately realized in the memory and the central processing unit (CPU). The necessity of communicating between them via the data bus, aka the von-Neumann bottleneck, not only decreases the computing speed, but also wastes significant amount of energy.

Spin-based technologies, nowadays mostly focusing on data storage such as the industrialized STT-MRAM and racetrack magnetic memories, are known for its high speed and non-volatility (thus energy efficiency). To overcome the von-Neumann bottleneck, we go beyond simply spin-based storage and introduce an in-memory computing structure of purely magnetic nature, which seamlessly integrates the memory and processing by employing the static magnetic texture for the data storage and its dynamical excitation – spin wave for processing. This in-memory computing architecture, utilizing the existing racetrack memories and synthetic antiferromagnet as its memory and processing components, is extremely simple in terms of hardware.

New mechanism for dissipative coupling

Thu, 28 Nov 2019 08:54:12 +0800

This work started last October during a KITS workshop held in Beijing. During the workshop, Prof. Can-ming Hu from University of Manitoba in Canada reported a non-typical energy level crossing [1], the so-called level attraction, in a YIG-cavity system. The level attraction is very different from the typical repulsive anti-crossing whenenver two energy levels intersecting each other.

Such level attraction behavior is very similar to the two precessing ferromagnets coupled via spin-pumping and spin-transfer torque studied more than a decade ago by Prof. Yaroslav Tserkovnyak [2]. Therefore, Prof. Xiao proposed after Hu’s talk that what Prof. Hu had observed shares the same physics as that seen in the ferromagnetic systems, and gave a simple coupled harmonic oscillator analog of the situation.

The crucial thing here is that the coupling must be of dissipative nature, where thes spin pumping/spin-transfer torque happens to be. In a coupled harmonic oscillator case, the dissipative coupling can be realized by friction or viscous force or via a third oscillator with high dissipation (see the figure on the right).

Spin X Workshop

Tue, 08 Oct 2019 08:54:12 +0800

The first Spin X Workshop was successfully held on Oct. 7th, 2019 in Baolong Hotel, Shanghai. Ten invited speakers from China, Japan, and The Netherlands shared their research ideas on spintronics ranging from spin mechanics, spin cavitronics, to topological effects in magnetic systems.

The workshop atmosphere is lively with prolonged hot discussions, and extended for more than 2 hours than planned.

We would like to thank all the speakers for their contributions to the successful gathering and exchange of ideas.

The detailed program of the workshop is shown below: