Pattern Classification Using Quantized Neural Networks for FPGA-Based Low-Power IoT Devices.

With the recent growth of the Internet of Things (IoT) and the demand for faster computation, quantized neural networks (QNNs) or QNN-enabled IoT can offer better performance than conventional convolution neural networks (CNNs). With the aim of reducing memory access costs and increasing the computation efficiency, QNN-enabled devices are expected to transform numerous industrial applications with lower processing latency and power consumption. Another form of QNN is the binarized neural network (BNN), which has 2 bits of quantized levels. In this paper, CNN-, QNN-, and BNN-based pattern recognition techniques are implemented and analyzed on an FPGA. The FPGA hardware acts as an IoT device due to connectivity with the cloud, and QNN and BNN are considered to offer better performance in terms of low power and low resource use on hardware platforms. The CNN and QNN implementation and their comparative analysis are analyzed based on their accuracy, weight bit error, RoC curve, and execution speed. The paper also discusses various approaches that can be deployed for optimizing various CNN and QNN models with additionally available tools. The work is performed on the Xilinx Zynq 7020 series Pynq Z2 board, which serves as our FPGA-based low-power IoT device. The MNIST and CIFAR-10 databases are considered for simulation and experimentation. The work shows that the accuracy is 95.5% and 79.22% for the MNIST and CIFAR-10 databases, respectively, for full precision (32-bit), and the execution time is 5.8 ms and 18 ms for the MNIST and CIFAR-10 databases, respectively, for full precision (32-bit).

A 65 nm Duplex Transconductance Path Up-Conversion Mixer for 24 GHz Automotive Short-Range Radar Sensor Applications.

A 24 GHz highly-linear upconversion mixer, based on a duplex transconductance path (DTP), is proposed for automotive short-range radar sensor applications using the 65-nm CMOS process. A mixer with an enhanced transconductance stage consisting of a DTP is presented to improve linearity. The main transconductance path (MTP) of the DTP includes a common source (CS) amplifier, while the secondary transconductance path (STP) of the DTP is implemented as an improved cross-quad transconductor (ICQT). Two inductors with a bypass capacitor are connected at the common nodes of the transconductance stage and switching stage of the mixer, which acts as a resonator and helps to improve the gain and isolation of the designed mixer. According to the measured results, at 24 GHz the proposed mixer shows that the linearity of output 1-dB compression point (OP1dB) is 3.9 dBm. And the input 1-dB compression point (IP1dB) is 0.9 dBm. Moreover, a maximum conversion gain (CG) of 2.49 dB and a noise figure (NF) of 3.9 dB is achieved in the designed mixer. When the supply voltage is 1.2 V, the power dissipation of the mixer is 3.24 mW. The mixer chip occupies an area of 0.42 mm2.

A Novel 65 nm Active-Inductor-Based VCO with Improved Q-Factor for 24 GHz Automotive Radar Applications.

The inductor was primarily developed on a low-voltage CMOS tunable active inductor (CTAI) for radar applications. Technically, the factors to be considered for VCO design are power consumption, low silicon area, high frequency with reasonable phase noise, an immense quality (Q) factor, and a large frequency tuning range (FTR). We used CMOS tunable active inductor (TAI) topology relying on cascode methodology for 24 GHz frequency operation. The newly configured TAI adopts the additive capacitor (Cad) with the cascode approach, and in the subthreshold region, one of the transistors functions as the TAI. The study, simulations, and measurements were performed using 65nm CMOS technology. The assembled circuit yields a spectrum from 21.79 to 29.92 GHz output frequency that enables sustainable platforms for K-band and Ka-band operations. The proposed design of TAI demonstrates a maximum Q-factor of 6825, and desirable phase noise variations of -112.43 and -133.27 dBc/Hz at 1 and 10 MHz offset frequencies for the VCO, respectively. Further, it includes enhanced power consumption that varies from 12.61 to 23.12 mW and a noise figure (NF) of 3.28 dB for a 24 GHz radar application under a low supply voltage of 0.9 V.

Design and Analysis of a Novel 24 GHz Up-Conversion Mixer with Improved Derivative Super-Position Linearizer Technique for 5G Applications.

A 24 GHz high linear, high-gain up-conversion mixer is realized for fifth-generation (5G) applications in the 65 nm CMOS process. The mixer's linearity is increased by applying an Improved Derivative Super-Position (I-DS) technique cascaded between the mixer's transconductance and switching stage. The high gain and stability of amplifiers in the transconductance stage of the mixer are achieved using novel tunable capacitive cross-coupled common source (TCC-CS) transistors. Using the I-DS, the third-order non-linear coefficient of current is closed to zero, enhancing the linearity. Additionally, a TCC-CS, which is realized by varactors, neutralizes the gate-to-drain parasitic capacitance (Cgd) of transistors in the transconductance stage of the mixer and contributes to the improvement of the gain and stability of the mixer. The measured 1 dB compression point OP1dB of the designed mixer is 4.1 dBm and IP1dB is 0.67 dBm at 24 GHz. The conversion gain of 4.1 dB at 24 GHz and 3.2 ± 0.9 dB, from 20 to 30 GHz is achieved in the designed mixer. Furthermore, a noise figure of 3.8 dB is noted at 24 GHz. The power consumption of the mixer is 4.9 mW at 1.2 V, while the chip area of the designed mixer is 0.4 mm2.