Nanoparticle network memory with reconfigurable synaptic interaction diafiltration pathways


The development of memory devices with the function of simultaneously processing and storing data is the need for efficient computing, and artificial synapses are devices proposed by people to achieve this goal, but the irreversible aging of these electrical devices will lead to a decline in their performance. Hoo-Cheol Lee et al. from Korea University’s Department of Physics demonstrated a nanoparticle network memory that uses a reconfigurable diafiltration path in a single silicon nanowire. Electrical and photonic control of the current percolation path enables simulation and reversible adjustment of persistent current levels, exhibiting memory behavior and current suppression in this single nanowire device. Electrical and photonic reconstruction of conductive paths in silicon nanoparticle networks will pave the way for the next generation of nanodevice technologies.


Simultaneous processing and storage of data in a single storage device is a need for efficient computing, and for this reason, artificial synaptic devices that control signal weights have been developed by mimicking synaptic behavior in biological systems. While the device’s array is capable of neuromorphic calculations, a single device can also form hybrid networks with biological neurons, enabling interaction and communication between the brain and computers. However, due to the irreversible aging caused by the evolution of the internal structure of the device, the performance degradation of these devices is inevitable.

On the other hand, photonic devices have been proposed to control current levels without causing device degradation. For example, photon-triggered transistors and atomically thin photonic transistors have successfully demonstrated high device performance, and photonic synapses for neuromorphic applications have also been demonstrated. However, photocurrent generation in these semiconductor devices is often used for current enhancement, and photonically suppressing current levels and switching analog conductance remains a challenge.

Hoo-Cheol Lee et al. from the Department of Physics at Korea University demonstrated a reconfigurable diafiltration path in nanoparticle network memory networks for synaptic interactions. The researchers used a single silicon nanowire with a solid core and porous shell segments. Electrical and photonic control of conductive paths in the silicon nanoparticle network of the nanowire shell efficiently adjusts persistent current levels in a simulated and reversible manner. In addition to memory behavior, photon habits in nanowire devices were demonstrated by abruptly breaking the current infiltration path under laser irradiation. In addition, using potential and habituation processes, the characteristics of nanoscale synaptic devices are demonstrated in this single nanowire memory. In particular, synaptic elimination is achieved in two adjacent devices connected to each other on a single nanowire by using photon habit as a cut-off switch under illumination. Electrical and photonic reconstruction of conductive pathways in silicon nanoparticle networks, as well as simulating synapses in nanowire memory, are expected to be key to next-generation nanodevice technologies.

Innovative research

Using a network of nanocrystals, the researchers can reversibly control persistent current paths without structural deformation, as shown in Figures 1a and 1b. Since many nanograins are interconnected, such a network has a high electrical resistance; However, the charge can form a current penetration path with a lower resistance. In fact, due to the self-capacitance nature of the nanograins, the charge is stored in the network. As the charge increases, the current flow process changes from an electron jump to a space charge limiting process (Figure 1a). It can be seen that the current of electron beating is much lower than that of space charge limiting, because the electrical connection in the diafiltration path is done by a network of charged nanoparticles, and the jumping process requires activation energy to overcome the Coulomb barrier. Therefore, by adjusting the electron runout and space charge limiting, it is feasible to simulate the control of persistent current levels.

Figure 1 depicts the flow of current in a nanograin network

Figure 1b The left panel depicts the current diafiltration path being blocked under illumination, and the right panel shows that as the incident light intensity increases, more current diafiltration paths are disconnected. Figure 1c is the calculated weight function, and Figure 1d and Figure 1e are the calculated parameter charge and total current as a function of the bias voltage, respectively.

The unique properties of nanoparticle networks can be achieved through porous silicon structures. The memory device is implemented by rationally designing a single silicon wafer with a solid core and a porous shell structure, as shown in Figure 2a. To fabricate a nanowire device with two structurally distinct segments, the researchers used metal-assisted chemical etching, as shown in the scanning electron microscope image of the nanowire structure in Figure 2b, with the left and right electrodes fabricated on long solid segments and short core/shell segments, respectively.

Figure 2 Memory behavior and photon habits in nanowire devices

Figure 2c shows the I-V curve of a measured double-scan mode nanowire device, Figure 2d depicts the measured current over time under laser irradiation, and Figure 2e shows the measured current vs. laser power in 2d.

Overall, electrical and photonic reconstruction of conductive paths in silicon nanoparticle networks opens up a new paradigm for next-generation nanodevice technologies. Since the smaller synaptic size reduces power consumption while increasing the integrated density, it will be of interest and interest to explore synaptic devices with synaptic density and energy efficiency comparable to the human brain using a single nanowire memory device.

The article, published in the journal Light: Science & Applications, titled “Nanograin network memory with reconfigurable percolation paths for synaptic interactions,” was published in the journal Light: Science & Applications, with Jungkil Kim and Hong-Gyu Park as corresponding authors. (Source: LightScience Applications WeChat public account)

Related paper information:‍-023-0‍1168-5

Special statement: This article is reproduced only for the need to disseminate information, and does not mean to represent the views of this website or confirm the authenticity of its content; If other media, websites or individuals reprint and use from this website, they must retain the “source” indicated on this website and bear their own legal responsibilities such as copyright; If the author does not wish to be reprinted or contact the reprint fee, please contact us.

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button