The thickening behavior of aluminum scandium nitride (Al0.88Sc0.12N) films grown on Si(111) substrates has been investigated experimentally using X-ray diffraction (XRD), transmission electron microscopy (TEM), atomic force microscopy, and residual stress measurement. Al0.88Sc0.12N films were grown with thicknesses spanning 14 nm to 1.1 um. TEM analysis shows that the argon sputter etch used to remove the native oxide prior to deposition produced an amorphous, oxygen-rich surface, preventing epitaxial growth. XRD analysis of the films show that the A1ScN(002) orientation improves as the films thicken and the XRD A1ScN(002) rocking curve full width half maximum decreases to 1.34 q for the 1.1 pm thick film. XRD analysis shows that the unit cell is expanded in both the a- and c-axes by Sc doping; the a-axis lattice parameter was measured to be 3.172 ± 0.007 A and the c-axis lattice parameter was measured to be 5.000 ± 0.001 A, representing 1.96% and 0.44% expansions over aluminum nitride lattice parameters, respectively. The grain size and roughness increase as the film thickness increases. A stress gradient forms through the film; the residual stress grows more tensile as the film thickens, from -1.24 GPa to +8.5MPa.
In this paper we describe a method for controlling both the residual stress and the through-thickness stress gradient of aluminum nitride (AlN) thin films using a multi-step deposition process that varies the applied radio frequency (RF) substrate bias. The relationship between the applied RF substrate bias and the AlN residual stress is characterized using AlN films grown on oxidized silicon substrates is determined using 100 nm-1.5 μm thick blanket AlN films that are deposited with 60-100 W applied RF biases; the stress-bias relationship is found to be well described using a power law relationship. Using this relationship, we develop a model for varying the RF bias in a series of discrete deposition steps such that each deposition step has zero average stress. The applied RF bias power in these steps is tailored to produce AlN films that have minimized both the residual stress and the film stress gradient. AlN cantilevers were patterned from films deposited using this technique, which show reduced curvature compared to those deposited using a single RF bias setting, indicating a reduction of the stress gradient in the films.
GaN-on-Si combines the wide bandgap advantages of GaN with the cost and scaling advantages of Si. Sputtered A1N is an attractive nucleation layer material because it reduces Al diffusion into the Si and eliminates a time-intensive preconditioning step in the GaN growth process, but is limited by the poor film quality of PVD A1N films deposited on Si substrates. Sputtering also offers a large degree of control over A1N film properties, including control of the intrinsic stress using substrate biasing. Doping the A1N films with Sc improves the lattice match to A1GaN and GaN films by expanding the a-axis and c-axis lattice parameters. A1N and A10.88Sc0.12N films have been grown on silicon, metal, and sapphire substrates and characterized for properties such as stress, grain size, roughness, and film orientation for use as nucleation layers for MOCVD GaN growth.
Resistive memory (ReRAM) shows promise for use as an analog synapse element in energy-efficient neural network algorithm accelerators. A particularly important application is the training of neural networks, as this is the most computationally-intensive procedure in using a neural algorithm. However, training a network with analog ReRAM synapses can significantly reduce the accuracy at the algorithm level. In order to assess this degradation, analog properties of ReRAM devices were measured and hand-written digit recognition accuracy was modeled for the training using backpropagation. Bipolar filamentary devices utilizing three material systems were measured and compared: one oxygen vacancy system, Ta-TaOx, and two conducting metallization systems, Cu-SiO2, and Ag/chalcogenide. Analog properties and conductance ranges of the devices are optimized by measuring the response to varying voltage pulse characteristics. Key analog device properties which degrade the accuracy are update linearity and write noise. Write noise may improve as a function of device manufacturing maturity, but write nonlinearity appears relatively consistent among the different device material systems and is found to be the most significant factor affecting accuracy. As a result, this suggests that new materials and/or fundamentally different resistive switching mechanisms may be required to improve device linearity and achieve higher algorithm training accuracy.
Resistive memory (ReRAM) shows promise for use as an analog synapse element in energy-efficient neural network algorithm accelerators. A particularly important application is the training of neural networks, as this is the most computationally-intensive procedure in using a neural algorithm. However, training a network with analog ReRAM synapses can significantly reduce the accuracy at the algorithm level. In order to assess this degradation, analog properties of ReRAM devices were measured and hand-written digit recognition accuracy was modeled for the training using backpropagation. Bipolar filamentary devices utilizing three material systems were measured and compared: one oxygen vacancy system, Ta-TaOx, and two conducting metallization systems, Cu-SiO2, and Ag/chalcogenide. Analog properties and conductance ranges of the devices are optimized by measuring the response to varying voltage pulse characteristics. Key analog device properties which degrade the accuracy are update linearity and write noise. Write noise may improve as a function of device manufacturing maturity, but write nonlinearity appears relatively consistent among the different device material systems and is found to be the most significant factor affecting accuracy. This suggests that new materials and/or fundamentally different resistive switching mechanisms may be required to improve device linearity and achieve higher algorithm training accuracy.