The operating principle of the proposed device is discussed focusing on how signal transmission through the ring type structure is available without any feed line between the antenna and the detector. From a fabricated ring-type FET-based monolithic antenna device, we demonstrated the greatly improved optical responsivity and reduced optical noise equivalent power, which are in comparable order with the reported state-of-the-art CMOS-based antenna integrated direct detectors.
Problem statement
Therefore, it is difficult to design the antenna and feed line with the input impedance corresponding to each MOSFET device pixel. Therefore, a careful MOSFET device design considering its input impedance and matching the antenna is one of the most important.
Motivation and scope
To solve the limitations in previous asymmetric FET-based designs and to realize a high-performance multi-pixel mmW plasmonic detector, we propose a new monolithic circular antenna-transistor in this thesis. In terms of the antenna, the proposed circular antenna structure is not selective for the polarization state of the incident wave.
Thesis outline
The new concept is to combine the functions of a plasmonic mmW detection and an antenna within the same structure. Furthermore, taking into account a practical mmW detection technology, the antennas must meet the requirements for high responsivity and detectivity, simple fabrication, identical photoresponse against different polarization angles, and the possibility of mmW detection over a wide bandwidth.
Principle of performance enhancement
The circular source and annular drain of the proposed structure can increase the charge asymmetry. Under the same asymmetry ratio (ηa = WD/WS) and the source-to-drain top edge distance (l) (in Figure 2.2), the annular channel has a higher electron density than the rod-shaped channel near the source side, since the source curvature in the annular structure has a more limited channel possible.
Resonance frequency
Equation (2.8) together with the boundary conditions (2.9) and (2.12) on the surface of the conducting pillars for the TM mode defines the problem of finding the eigenvalue. The reflection from different lines with matching phases can create a complete reflection of the wave from a set of metal patches, thus creating a band gap.
Influence of substrate thickness and its modes
Using an internal Matlab code to plot the scattering characteristics of the 2D periodic array in air (r = 1) and silicon (r = 11.9) medium, we can determine the value of r/a that initializes the resonance frequency at Γ -point (cutoff) frequency) of the first dispersion curve (air, tSi = 0 µm). Furthermore, the substrate conditions affect the performance of the detector, we could not consider the effects in the actual design process due to limitations of clean room fabrication facilities.
Available power on transistor-antenna
The available power to the detector element antenna pixel was estimated by the following path-loss formula: where Ptis is the source power fed to the horn antenna, α power ratio. The power ratio α is the ratio of beam power in detector's active area to total beam power at the detector plane, or it indicates how much of the total power is captured by the detector active area. α can be calculated based on Gaussian field distribution of the paraxial Gaussian beam. Where we assume that the source power P is equal to the total beam power at detector level (z=f arfield distance) by neglecting the atmospheric losses.
As seen in the above expression, the power ratio basically takes into account the free space losses and the gain of the transmitting horn antenna. The gain of the antenna element can be defined as the ratio of the intensity, in a given direction (Urad(θ, φ)), to the radiation intensity that would be obtained if the power received by the antenna was radiated isotropically [64] . The radiation intensity corresponding to isotropically radiated power is equal to the input power (Pin) delivered to the antenna divided by 4π.
Monolithic circular transistor-antenna device
In this chapter, we introduce the design and characterization of a circular transistor antenna structure, focusing on its effect on improved detector performance results in terms of the plasmonic and electromagnetic aspects. Due to a manufacturing limit, the substrate thickness tSi of the transistor antenna structure must be greater than 0.1λd, whereλ is the wavelength within the dielectric substrate. Since we design the antenna on a thick and conductive silicon substrate, the resonance frequency estimation requires a parametric study supported by EM simulation data, which is discussed in Sec.
Design and characterization
A simplified layout of a pixel of the ring-type FET-based monolithic circular antenna is shown in Fig. We used the High Frequency Structure Simulator (HFSS) [68] to simulate the electromagnetic responses of the circular antenna. b) Side view of ring-type FET-based monolithic circular antenna. The resonance frequency depends on pixel size, dielectric thickness tSi and the electrical permittivity of the substrate.
Although the mathematical description of the relationships between the performance and the design variables is difficult, we can simplify the problem by considering the 2-D periodic array of circular metal patches with zero substrate thickness for the first step estimation of the resonance frequency. For the system with the 2-D periodic array of circular metal patches with kz = 0, we can obtain band gap diagrams as discussed in Sec. The estimation of the range and the initial ascertainment of the resonant frequency enable the reduction of the design effort before the other design parameters are determined.
Results and discussion
The reactive component of the impedance is about zero at the resonance frequency, where the power loss of the monolithic circular antenna becomes negligible. The measurement results of the ring-type FET-based monolithic circular antenna and the in-plane patch reference antenna with feeder lines are shown in Fig. The dc output ∆u = 46.9 µV is greatly increased from 5 µV of the same patch antenna reference sample ηa= 10, as shown in Fig.
To verify our design approach, we also measured and compared the performance values of the ring-type FET-based monolithic circular antenna for two silicon substrate thicknesses. The increase in the thickness of silicon substrate degrades the performance of the ring-type FET detector as additional substrate modes are generated. The measured bandwidth of the entire detector response covers more than half of the antenna bandwidth features in Fig.
EM structure design through Bayesian Learning
Our proposed BCL uses data-embedded statistical clique to find optimal design parameters [76–79] . Our algorithm reduces the complexity of generating networks by partitioning into K clusters and building a statistical clique. We denote the band stiffness matrix with the bandwidth W, the number of nodes N, and the number of elements E,M is the network input [82, 83].
The computational speedup with our algorithm is approximately O(K×s×tM . dist)), where O(E) is the complexity of generating the mesh, K is the number of clusters, s is the sample size, tdist is the time to calculate pairwise distances. The x-axis is the distance between two neighboring nodes of our statistical clique, and the y-axis is the number of neighboring nodes. The Euclidean threshold is the maximum limit on the x-axis above which the probability of obtaining statistical cliques is small.
Inverse design of EM structure with deep learning
A theoretical area of interest is the expressiveness of sparsely connected deep neural networks for inverse planning. From a practical point of view, generative adversarial networks (GANs) are used to inversely design a metamaterial structure as shown in the figure. The basic agent-environment framework consists of: (i) agent (ii) environment (iii) goals, as shown in fig.
We have mentioned an action space and similarly the environment sends signals to the agent with symbols from the perception space P. The reward space, denoted by R, denoted as:. We applied the above learning method to the inverse design of a metamaterial structure (in Figure 4.19) based on the desired set of frequency characteristics. Therefore, we developed an inverse design approach that finds suitable geometric values of transmission lines based on the required transmission characteristics.
Inverse design of transistor-antenna through Deep LearningLearning
Histogram training geometry parameter plots of n×nmetamaterial prototype structure. a) structure lengtha(b) gap width c(c) substrate lengthgx (d) line width w. We validated our inverse design via scatter plots with the value of the root mean square error (RMSE) in Fig. We plot our machine learning predicted parameters versus experimental parameters of the relative permittivity of the substrate r, dielectric thickness tSi, drain radius rdr, used in the corresponding simulation of the channel and we calculate the root mean square error as given in Fig.
The observed mismatch is largely due to training our model with fewer input samples of absorption spectra and partly due to the fact that the estimated function is not injective. In the future, we want to train our model with a larger number of input datasets to improve the root mean square errors.
Conclusion
We classify our data into three ranges and conduct our research to find range-specific parameters. Our learning algorithm is scalable and works on any general electromagnetic structure for automated design. We have placed a limit on the computational complexity of our method and discuss the trade-off between complexity and uncertainty.
The result of metamaterial prototype shows that the proposed learning algorithm can facilitate the design of MTMs widely with high accuracy. Our BCL allows us to reuse learning parameters from trained EM dataset to new EM dataset with small changes. While in another case, we have proposed a reverse design approach to find geometric parameters of EM structures from frequency characteristics.
Future Work
Gonzales-Valdes et al., "Millimeter wave imaging architecture for on-the-move whole body imaging,"IEEE Transactions on Antennas and Propagation, vol. Raptis, "Millimeter-Wave Radar Sensing of Airborne Chemicals," IEEE Transactions on Microwave Theory and Techniques, vol. Croskey et al., "The Millimeter Wave Atmospheric Sounder (MAS): A Shuttle-Based Remote Sensing Experiment,"IEEE Transactions on Microwave Theory and Techniques, vol.
Myers et al., “Antenna-Coupled Bolometers for Millimeter Waves”, IEEE Transactions on Applied Superconductivity, vol. Deutsch et al., “Frequency-Dependent Loss on High-Performance Interconnections”, IEEE Transactions on Electromagnetic Compatibility, vol. Gentile et al., “Silicon-Filled Rectangular Waveguides and Frequency Scanning Antennas for mm-Wave Integrated Systems”, IEEE Transactions on Antennas and Propagation, vol.
