Synaptic transistor; all-optical switches; thermal in electronic/photonic integration.
Researchers from Northwestern University, Boston College, and MIT developed a synaptic transistor that simultaneously processes and stores information similar to the human brain. The team said the transistor goes beyond simple machine learning tasks to categorize data and is capable of performing associative learning.
The new device is stable at room temperatures. It also operates at fast speeds, is energy efficient, and retains stored information when power is removed.
The researchers combined two different types of atomically thin materials: bilayer graphene and hexagonal boron nitride. When stacked and purposefully twisted, the materials formed a moiré pattern. By rotating one layer relative to the other, the researchers could achieve different electronic properties in each graphene layer even though they are separated by only atomic-scale dimensions. With the right choice of twist, researchers harnessed moiré physics for neuromorphic functionality at room temperature.
“With twist as a new design parameter, the number of permutations is vast,” said Mark C. Hersam, professor of materials science and engineering, medicine, and chemistry at Northwestern, in a statement. “Graphene and hexagonal boron nitride are very similar structurally but just different enough that you get exceptionally strong moiré effects.”
To test the transistor, the team trained it to recognize similar, but not identical, patterns. First the researchers showed the device one pattern: 000 (three zeros in a row). Then, they asked the AI to identify similar patterns, such as 111 or 101. “If we trained it to detect 000 and then gave it 111 and 101, it knows 111 is more similar to 000 than 101,” Hersam explained. “000 and 111 are not exactly the same, but both are three digits in a row. Recognizing that similarity is a higher-level form of cognition known as associative learning.”
In experiments, the new synaptic transistor successfully recognized similar patterns, displaying its associative memory. It was also able to demonstrate associative learning when given incomplete patterns. [1]
Researchers from Argonne National Laboratory and Purdue University built an all-optical switch out of two different materials, each with a different switching time. One material, aluminum-doped zinc oxide, has a switching time in the picosecond range, while the other material, plasmonic titanium nitride, has a switching time more than a hundred times slower, in the nanosecond range.
“Previous iterations of optical switches had fixed switching times that were ‘baked in’ to the device upon its fabrication,” said Soham Saha, a postdoctoral fellow at Argonne, in a statement. Saha added that the difference in switching times between the two metal components means that the switch can be more flexible and used to both transmit data quickly while also storing it effectively. “The bimetallic nature of the switch means that it can be used for multiple purposes depending on the wavelength of the light that you use. When you want slower applications, like memory storage, you switch with one material; for faster applications, you switch with the other one. This capability is new.”
In the experimental configuration, the materials of the switch function as light absorbers or reflectors, depending on the wavelength of operation. When they are switched on by a light beam, they switch state. [2]
Researchers from KU Leuven and imec investigated the thermal impact of 3D integration of electronic chips on top of photonic chips. The photonic chip consisted of an array of ring modulators, which are temperature sensitive and can require active thermal stabilization with an integrated heater in demanding environments.
The team measured the heater efficiency of the ring modulators experimentally before and after flip-chip bonding of the EIC on the PIC. They found a relative loss of -43.3% in efficiency, a significant impact. “Furthermore, 3D finite element simulations attributed this loss to heat spreading in the EIC. This heat spreading should be avoided, because in the ideal case all heat that is generated in the integrated heater is contained close to the photonic device. The thermal crosstalk between the photonic devices also increased by up to +44.4% after bonding the EIC, which complicates the individual thermal control,” they noted in a release.
The researchers also found that by increasing the spacing between microbumps and the photonic device, and by decreasing the interconnect linewidth, the thermal penalty of 3D integration can be minimized. [3]
References
[1] Yan, X., Zheng, Z., Sangwan, V.K. et al. Moiré synaptic transistor with room-temperature neuromorphic functionality. Nature 624, 551–556 (2023). https://doi.org/10.1038/s41586-023-06791-1
[2] Saha, S., Diroll, B.T., Ozlu, M.G. et al. Engineering the temporal dynamics of all-optical switching with fast and slow materials. Nat Commun 14, 5877 (2023). https://doi.org/10.1038/s41467-023-41377-5
[3] David Coenen, Minkyu Kim, Herman Oprins, Yoojin Ban, Dimitrios Velenis, Joris Van Campenhout, Ingrid De Wolf, “Thermal modeling of hybrid three-dimensional integrated, ring-based silicon photonic–electronic transceivers,” J. Optical Microsystems 4(1) 011004 (6 December 2023) https://doi.org/10.1117/1.JOM.4.1.011004
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply