Both β-Ga₂O₃ and wurtzite AlN have wide bandgaps of 4.5–4.9 and 6.1 eV, respectively. We calculated the in-plane lattice mismatch between the (−201) plane of β-Ga₂O₃ and the (0002) plane of AlN, which was found to be 2.4%. This is the smallest mismatch between β-Ga₂O₃ and binary III-nitrides which is beneficial for the formation of a high quality β-Ga₂O₃/AlN heterojunction. However, the valence and conduction band offsets (VBO and CBO) at the β-Ga₂O₃/AlN heterojunction have not yet been identified. In this study, a very thin (less than 2 nm) β-Ga₂O₃ layer was deposited on an AlN/sapphire template to form the heterojunction by pulsed laser deposition. High-resolution X-ray photoelectron spectroscopy revealed the core-level (CL) binding energies of Ga 3d and Al 2p with respect to the valence band maximum in individual β-Ga₂O₃ and AlN layers, respectively. The separation between Ga 3d and Al 2p CLs at the β-Ga₂O₃/AlN interface was also measured. Eventually, the VBO was found to be −0.55 ± 0.05 eV. Consequently, a staggered-gap (type II) heterojunction with a CBO of −1.75 ± 0.05 eV was determined. The identification of the band alignment of the β-Ga₂O₃/AlN heterojunction could facilitate the design of optical and electronic devices based on these and related alloys.

The performance of a biologically plausible spiking neural network (SNN) largely depends on the model parameters and neural dynamics. This article proposes a parameter optimization scheme for improving the performance of a biologically plausible SNN and a parallel on-field-programmable gate array (FPGA) online learning neuromorphic platform for the digital implementation based on two numerical methods, namely, the Euler and third-order Runge–Kutta (RK3) methods. The optimization scheme explores the impact of biological time constants on information transmission in the SNN and improves the convergence rate of the SNN on digit recognition with a suitable choice of the time constants. The parallel digital implementation leads to a significant speedup over software simulation on a general-purpose CPU. The parallel implementation with the Euler method enables around $180\times $ ( $20\times $ ) training (inference) speedup over a Pytorch-based SNN simulation on CPU. Moreover, compared with previous work, our parallel implementation shows more than $300\times $ ( $240\times $ ) improvement on speed and $180\times $ ( $250\times $ ) reduction in energy consumption for training (inference). In addition, due to the high-order accuracy, the RK3 method is demonstrated to gain $2\times $ training speedup over the Euler method, which makes it suitable for online training in real-time applications.

The spontaneous polarization (SP) and piezoelectric (PZ) constants of B_xAl_1-xN and B_xGa_1-xN (0 ≤ x ≤ 1) ternary alloys were calculated with the hexagonal structure as reference. The SP constants show moderate nonlinearity due to the volume deformation and the dipole moment difference between the hexagonal and wurtzite structures. The PZ constants exhibit significant bowing because of the large lattice difference between binary alloys. Furthermore, the PZ constants of B_xAl_1-xN and B_xGa_1-xN become zero at boron compositions of ∼87% and ∼74%, respectively, indicating non-piezoelectricity. The large range of SP and PZ constants of B_xAl_1-xN (BAlN) and B_xGa_1-xN (BGaN) can be beneficial for the compound semiconductor device development. For instance, zero heterointerface polarization $\Delta P$ can be formed for BAlN and BGaN based heterojunctions with proper B compositions, potentially eliminating the quantum-confined Stark effect for c-plane optical devices and thus removing the need of non-polar layers and substrates. Besides, large heterointerface polarization $\Delta P$ is available that is desirable for electronic devices.

To solve real-time challenges, neuromorphic systems generally require deep and complex network structures. Thus, it is crucial to search for effective solutions that can reduce network complexity, improve energy efficiency, and maintain high accuracy. To this end, we propose unsupervised pruning strategies that are focused on pruning neurons while training in spiking neural networks (SNNs) by utilizing network dynamics. The importance of neurons is determined by the fact that neurons that fire more spikes contribute more to network performance. Based on these criteria, we demonstrate that pruning with an adaptive spike count threshold provides a simple and effective approach that can reduce network size significantly and maintain high classification accuracy. The online adaptive pruning shows potential for developing energy-efficient training techniques due to less memory access and less weight-update computation. Furthermore, a parallel digital implementation scheme is proposed to implement spiking neural networks (SNNs) on field programmable gate array (FPGA). Notably, our proposed pruning strategies preserve the dense format of weight matrices, so the implementation architecture remains the same after network compression. The adaptive pruning strategy enables 2.3× reduction in memory size and 2.8× improvement on energy efficiency when 400 neurons are pruned from an 800-neuron network, while the loss of classification accuracy is 1.69%. And the best choice of pruning percentage depends on the trade-off among accuracy, memory, and energy. Therefore, this work offers a promising solution for effective network compression and energy-efficient hardware implementation of neuromorphic systems in real-time applications.

Rapid advancements in 5G communication and the Internet of Things have prompted the development of cognitive radio sensing for spectrum monitoring and malicious attack detection. An end-to-end radio classification system is essential to realize efficient real-time monitoring at the edge. This work presents an end-to-end neuromorphic system enabled by an efficient spiking neural network (SNN) for radio classification. A novel hardware-efficient spiking encoding method is proposed leveraging the Sigma-Delta modulation mechanism in analog-to-digital converters. It requires no additional hardware components, simplifies the system design, and helps reduce conversion latency. Following a designed hardware-emulating conversion process, the classification performance is verified on two benchmark radio modulation datasets. A comparable accuracy to an artificial neural network (ANN) baseline with a difference of 0.30% is achieved on the dataset RADIOML 2018 with more realistic conditions. Further analysis reveals that the proposed method requires less power-intensive computational operations, leading to 22× lower computational energy consumption. Additionally, this method exhibits more than 99% accuracy on the dataset when the signal-to-noise ratio is above zero dB. The SNN-based classification module is realized on FPGA with a heterogeneous streaming architecture, achieving high throughput and low resource utilization. Therefore, this work demonstrates a promising solution for constructing an efficient high-performance end-to-end radio classification system.