In this paper, to overcome the difficulty of a narrow beam not being able to cover all users in the same cluster, we present a beamwidth adjustment-based scheme for non-orthogonal multiple access (NOMA) in hybrid millimeter wave (mmWave) communication systems. Our approach addresses the limitation of narrow analog beams by using a new hierarchical codebook based on NOMA to widen the analog beamwidth to cover all users of the same NOMA cluster. Simulation results demonstrate that our algorithm outperforms similar approaches in terms of performance.

This work presents the design and implementation of an 8-bit current-mode digital-to-analog converter (DAC) with a built-in calibration mechanism for active-matrix organic light-emitting diode (AMOLED) display drivers. Due to variations in thin-film transistor threshold voltage and mobility, AMOLED panels suffer from current non-uniformity and luminance degradation, necessitating accurate current sources with high channel-to-channel uniformity. The proposed current-mode DAC employs a channel calibration technique based on a two-step coarse/fine adjustment, utilizing a successive approximation register control loop and current comparator with correlated double sampling to mitigate offset. A 52-channel prototype was fabricated in a 0.18-μm CMOS process, occupying 0.00918 mm² per channel and consuming 7.2 μW. Measurement results demonstrate a maximum differential nonlinearity of 1.3 LSB and integral nonlinearity of 1.8 LSB.

A P-channel RC-LIGBT is required to configure a common-emitter complementary inverter instead of the conventional N-N inverter consisting of N-channel RC-LIGBTs on the high and the low side. However, the static breakdown voltage and the conductivity of the P-channel is low in comparison with those of the N-channel. To improve the conductivity and the static and dynamic breakdown voltages, the novel trench type P-channel RC-LIGBT with the dual drift layer on SOI is introduced in addition to the N-channel. It is demonstrated that the newly proposed P-channel RC-LIGBT can improve the conductivity and secure the static breakdown voltage more than 200 V by the trench gate with rounded trench edges and the dual drift structure consisting of a P drift region with a P top layer and an N drift region. Moreover, the reverse conduction is achieved by two PN diodes formed in series by shorting the floating P and N bases of the embedded thyristor through an ohmic contact with metal. This RC-LIGBT structure can avoid the snapback in the forward conduction. With the proposed P-channel including the N-channel RC-LIGBTs on SOI, it is verified that the common-emitter complementary half bridge inverter can operate without the penetration current and it is found that the additional P top layer on the P drift can suppress the current oscillation observed in the waveform of the complementary inverter during the turn-off of the P-channel having the dual drift structure.

This paper reports on a transimpedance amplifier (TIA) designed specifically for room-temperature mercury cadmium telluride (HgCdTe) photovoltaic diodes. Based on auto-zeroing technology, this design successfully reduces the op amp's equivalent input offset voltage to 7.6 μV, effectively suppressing the detector's dark current induced by the offset voltage, thus resolving a core challenge in high-precision infrared readout circuits near zero bias. Built on a 0.18 μm TSMC, 3.3 V CMOS process, this technology uses an auxiliary input to correct the offset voltage and employs linear components and a negative feedback loop to suppress nonlinear distortion. Furthermore, the design ensures a highly symmetrical layout of the op amp and guard rings within the layout, optimizing the system structure and reducing offset and noise. The on-chip area is approximately 0.16 mm². Under 3.3 V supply voltage and room temperature, the amplifier achieves an effective input-referred noise of 8.3 μV within a 1-400 Hz bandwidth. The offset voltage before auto-zero is 480 μV, and it decreases by 98.4% after auto-zeroing. A photovoltaic-type HgCdTe photodiode is connected across the auto-zero operational amplifier with added bias voltage, resulting in an output swing improvement of over 0.55 V. The design exhibits good linearity and effectively suppresses the increase in dark current caused by the offset voltage in the HgCdTe photodiode. A comparison between simulation and measurement results confirms the design's accuracy and validates the effectiveness of the auto-zero technique under real-world conditions.

A sensorless control method based on a super-twisting sliding mode observer (STSMO) and a second-order generalized integrator phase-locked loop (SOGI-PLL) with fuzzy control is proposed. The fuzzy control algorithm is integrated into the SOGI-PLL to reduce chattering and improve the accuracy of the rotor position estimation of the permanent magnet synchronous motor (PMSM). Firstly, the basic principles and mathematical models of the STSMO are explained in detail. Secondly, the basic concepts of the SOGI-FLL and the PLL are introduced, along with their applications in sensorless control. Then, based on the proposed method, a PMSM control strategy is developed following system modeling, observer design, and control strategy optimization. The effectiveness of the proposed method is verified through a comparative analysis between the optimized control strategy and the initial control strategy. The experimental results demonstrate that the proposed STSMO-SOGI-PLL method reduces velocity fluctuations to 40r/min and effectively eliminates steady-state estimation errors.

ReRAM-based Compute-in-Memory (CiM) architectures can accelerate AI workloads. However, device-level variations in ReRAM distort the distribution of multiply-accumulate (MAC) values, thereby degrading computation accuracy. This paper proposes a non-uniform quantization (NUQ) scheme optimized by genetic algorithms (GA) to reduce MAC readout errors while maintaining high computational efficiency. The proposed NUQ method adapts to resistance variations and signal aggregation effects. Under multi-weight and aging scenarios, it effectively mitigates signal overlap and improves readout accuracy by 60%.

This study investigates amorphous gallium oxide (a-GaO_x) thin-film UV-C photodetectors fabricated on synthetic amorphous quartz glass substrates using fine channel mist chemical vapor deposition (CVD) at 260°C. Post-deposition annealing (PDA) was performed at temperatures up to 800°C. X-ray diffraction confirmed that the films maintained their amorphous structure even after high-temperature PDA, in contrast to sapphire substrates, where crystallization and Al atom diffusion are typically observed. Optical measurements revealed that the optical bandgap (E_g,^opt) remained nearly constant after PDA, indicating that the quartz substrates effectively suppressed bandgap widening. PDA significantly reduced the dark current by more than two orders of magnitude, enhanced the UV-C wavelength selectivity, and suppressed persistent photoconductivity by shortening the response time. Device performance metrics, including responsivity (R), detectivity (D^*), external quantum efficiency (EQE), and photo-to-dark current ratio (PDCR), were markedly improved, with the best results obtained for films treated at 800°C. These findings highlight the efficacy of combining quartz substrates with PDA to realize high-performance a-GaO_x UV-C photodetectors.

This study proposes a novel analog integrated circuit to solve the two-armed bandit (2AB) problem, which is a fundamental decision-making task in reinforcement learning. The 2AB problem involves repeatedly selecting the better of two uncertain options to maximize cumulative rewards. While conventional implementations have relied on software or digital circuits, limitations in clock frequency and memory access latency have constrained further improvements in processing speed. To address this issue, we propose a circuit architecture that leverages the physical characteristics of analog components for high-speed operation. The proposed circuit is fabricated using TSMC 180-nm CMOS process and integrates a dynamic comparator along with a charge pump-based history accumulation circuit (HAC). This configuration enables real-time adaptation to changes in reward probability based on the Thompson sampling strategy. In addition, a pulse generation circuit is introduced to clearly separate the phases of selection and learning, thereby suppressing timing mismatches and hazards to ensure stable operation. The prototype chip was evaluated under various reward probability settings and with a clock frequency of up to 1.0 GHz. Experimental results confirmed correct learning behavior at frequencies up to 840 MHz. However, a certain degree of selection imbalance caused by asymmetry in the charge and discharge operations of the HAC was observed. Future work will focus on improving the charge/discharge balance of the HAC and optimizing the comparator design to further enhance performance.

A building block type computing system is constructed by stacking various chips three-dimensionally using an inductively coupled wireless communication interface TCI (Through Chip Interface). Vertical buses using TCI have the benefit of forming a multi-drop shared bus among multiple chips. However, it requires magnetic flux to pass between multiple chips, so the diameter of the coil needs to be increased. In a typical shared bus, it is common to provide dedicated signals for arbitration. However, in the case of TCI, the area cost is too high to use a large-diameter coil exclusively for arbitration. Therefore, arbitration must be performed only by detecting data collisions in the TCI bus, and this point is similar to arbitration methods for other wireless buses, which perform arbitration based only on signal collisions. However, TCI is limited in the small scale of the chip stack, and there is a synchronization signal between chips. We take advantage of this feature and use chip IDs to sort the order of data transmission by each chip when a collision occurs within the TCI bus. We propose a Collision Sort method to do this. The proposed method was evaluated with a cycle-level simulation, and it was confirmed that the performance was close to that of the Ideal Shared Bus with a dedicated arbitration bus.

Heterogeneous systems equipped with GPUs on CPUs have become common in the fields of HPC and AI. NVIDIA GPUs are widely used, and many applications are written in CUDA. The SIMT programming model, such as CUDA, differs from the SIMD programming model used in CPUs, which means it often cannot be executed directly on CPUs. Therefore, users must select programming models suited to various systems and reimplement applications using them. In this paper, we propose a transpiler that translates CUDA host and kernel code into OpenMP code. By translating the source code from the SIMT programming model to the SIMD programming model, we aim to enhance the performance portability of applications and enable users to optimize the translated code manually. To demonstrate the utility of the proposed method, we performed performance comparisons using Rodinia Benchmark and NAS Parallel Benchmark among the translated code, the original OpenMP implementation, and prior work, namely MCUDA, Polygeist, and CuPBoP. The translated code showed performance comparable to the original OpenMP implementation in many benchmarks, although there were a few benchmarks where the performance significantly decreased. We also provided an analysis of the causes. This paper highlights the improvement in performance portability of applications written in CUDA and indicates issues for achieving further performance improvements.

Analog circuit sizing remains a challenging task owing to nonlinear behavior, complex trade-offs, sensitivity to PVT variations, and the high computational cost of transistor-level simulations. To address these issues, we propose a simulation-efficient analog circuit sizing approach that adaptively adjusts simulation fidelity according to the design stage. The method operates in of two stages: an initial sampling phase using fast operating point analysis to identify high-quality design candidates, followed by a Bayesian optimization phase that refines these candidates using high-fidelity simulations. A Sobol sequence is employed to generate uniformly distributed design points, and domain knowledge is integrated into the low-fidelity evaluations to guide sampling toward feasible and high-performing regions. Experimental validation on an operational amplifier designed using the TSMC 180nm CMOS process shows that the proposed method reduces the number of optimization iterations required to find a feasible solution to 40% of that of the conventional approach. Furthermore, the proposed method yields designs with improved DC gain and lower power consumption in constrained optimization scenarios, confirming the benefit of domain-informed sampling. These results suggest that incorporating domain knowledge into fidelity-adaptive optimization is a promising direction for scalable and interpretable analog circuit design automation.

Terrestrial single-event transients (SETs) are mainly induced by secondary ions of neutron-induced nuclear reactions, which do not cross the sensitive drain in most cases. Technology computer-aided design (TCAD) simulations show that in those cases, both diffusion-collection type and burst-plateau-type current waveform could occur in nanoscale complementary metal-oxide-semiconductor (CMOS) technology. “Secondary funneling” effect has been observed when the secondary ions are not incident on the drain. The necessity of including recombination effect in diffusion-collection model is identified, and an equivalent circuit model (ECM) is established to describe SETs induced by secondary ions with short distance to the drain. Both models are verified by TCAD simulations.

This paper proposes an extremely low-voltage flip-flop (ELVFF) consisting of recursive stacking body-bias (RSBB) logic gates. The ELVFF has the capability of operating at extremely low supply voltages. The ELVFF is based on a conventional NAND latch-based flip-flop (NLFF) and consists of three-times RSBB NANDs (3RSBB-NANDs) and one-time RSBB inverters (1RSBB-INVs). These techniques will provide a promising solution for achieving ELV operation. These techniques enhances both the voltage swing and voltage gain of logic gates, enabling the ELVFF to operate at extremely low supply voltages. Simulation results in a 0.18-μm CMOS process technology with a deep n-well option showed that the proposed ELVFF can operate at an extremely low supply voltage, below 50 mV. Measurement results also demonstrated that the ELVFF can latch and retain the correct logic data with a voltage swing of 27 mV at a 44-mV power supply. The proposed ELVFF is suitable for sub-100-mV ELV applications, such as energy harvesting, at the cost of an increased number of transistors, area, power, and delay time.

This paper presents ring oscillators (ROSCs) capable of operating at extremely low supply voltages. For inverters to have extremely low-voltage operation, we developed dedicated low-voltage inverters: a recursive stacking inverter (RS-INV) and a recursive stacking body-bias inverter (RS-BBI). The RS-INV is based on a conventional stacked inverter and consists of additional inverters stacked on the top and bottom of the inverters recursively. To further enhance low-voltage operation, we propose an RS-BBI that combines the RS-INV with the body-bias technique. Measurement results indicated that our ROSCs using RS-BBIs were able to operate at a lower supply voltage V_DD as the number of stacking inverters increased. The measured lowest V_DD,min was 26 mV even though not all chips were able to oscillate. All chips oscillated successfully at 30 mV.

This paper investigates the hybrid rotor design integrating consequent-pole permanent magnets (CPMs) and spoke-type permanent magnets (SPMs), designated as the CS-FMPM machine, and optimizes its configuration. To maximize its electromagnetic performance, the optimization method integrating improved multi-objective coati optimization algorithm (IMOCOA) with response surface methodology (RSM) is performed. Initially, the average torque, torque ripple, and cogging torque are defined as optimization objectives, while critical structural parameters are selected as optimization variables. Subsequently, the comprehensive sensitivity analysis method is employed to categorize the design variables into two hierarchical levels: sensitive level 1 and nonsensitive level 2. The generalized regression neural network (GRNN) model and the simple RS model are then utilized to approximate the optimization objectives for level 1 and level 2 variables. Furthermore, the IMOCOA and RSM are sequentially applied to optimize the level 1 and level 2 variables. Finally, finite element analysis (FEA) models of both the initial and optimized CS-FMPM machines are constructed, and the electromagnetic performances are comparatively analyzed. It demonstrates that the optimized CS-FMPM machine exhibits enhanced average torque and improved PM utilization efficiency, accompanied by the significant reduction in torque ripple and cogging torque. The proposed multi-objective optimization methodology for the CS-FMPM machine enables the rapid and efficient derivation of optimal design solutions. This approach not only elevates the machine's overall operational performance but also substantially shortens the optimization cycle, thereby validating the effectiveness and advantages of the proposed optimization methodology.

To increase the degree of integration of silica-based planar lightwave circuits for optical communications, we investigated the stacking of waveguides in layers by using flame hydrolysis deposition (FHD). Here, although FHD, a glass deposition method used in the manufacture of silica-based planar lightwave circuits, has high manufacturability, it requires a high-temperature heat treatment to consolidate glass materials. To fabricate stacked waveguides, it is necessary to lower the heat-treatment temperature when forming the upper waveguide layers to prevent deformation of the lower waveguide layers formed earlier. However, using different heat-treatment temperatures requires different amounts of dopants in each glass layer. This leads to different internal stresses induced in the various waveguide layers, leading polarization dependence of the equivalent refractive index via the photo-elastic effect. In this paper, we describe the fabrication and stress-induced birefringence in stacked silica-based planar lightwave circuits and discuss in detail fundamental optical circuit elements including the interlayer optical directional couplers, interlayer Mach-Zenhder interferometers, and array waveguide gratings that can be formed in such layers. In addition, the methods to eliminate polarization dependence are proposed for each element.

The high-speed switching processes in power electronic devices cause severe conducted electromagnetic interference (EMI), threatening system stability. Traditional methods struggle to accurately predict the insertion loss (S12) of T-type EMI filters, as they fail to effectively characterize high-frequency parasitic parameters and component mutual coupling effects, while relying on empirical 3D electromagnetic simulation proves inefficient. This paper proposes a data-driven approach based on Gaussian Process Regression (GPR) for predicting the S12 of T-filters across the full frequency band (150 kHz-30 MHz). This method treats the filter as a "black-box", learning the complex nonlinear mapping relationship between key filter parameters (inductance L, capacitance C, component spacing d) and S12 from measured data. The study utilized a Vector Network Analyzer (VNA) to experimentally measure 75 sets of data from LCL-type filters with different parameter combinations, employing 65 sets for training the GPR model and 10 sets for validation. Results demonstrate that the GPR model effectively captures component parasitic effects and mutual coupling influences. It exhibits high prediction accuracy across the divided sub-bands - the low-frequency band (150kHz-800kHz), mid-frequency band (800kHz-5MHz), and high-frequency band (5MHz-30MHz) - and provides uncertainty estimates for predictions. This method significantly enhances the efficiency and accuracy of EMI filter performance prediction, For the tested T-type filter, 3D modeling and simulation takes 20 minutes, while Gaussian process regression only requires 10 minutes. The method offers reliable guidance for rapid design and parameter optimization.

In order to improve the output of an enzyme-based biofuel cell (BFC) that uses plant as fuel, an enzyme of pectinase was used with cellulase, a cellulose-decomposing enzyme, in this study. Pectinase breaks down pectin, a tissue that holds cellulose fibers together in plants. Therefore, pectinase was used to promote the decomposition of cellulose. Finely chopped plant leaf aqueous solution was used as fuel. The plant was converted into glucose using the above two decomposing enzymes. First, the optimum conditions for pectinase, such as concentration, decomposition time, and reaction temperature, were determined by CV method. Next, the output (power density) was measured and compared with other fuels, such as glucose and cellulose nanofiber (CNF) solutions. The addition of pectinase improved the power density by approximately three times (52 μW/cm²), which was almost equivalent to the power output of the optimally concentrated glucose solution fuel (46 μW/cm²).

Realization of a highly efficient power combining technique is essential for increasing the output power from a monolithic microwave integrated circuit (MMIC) power amplifier (PA). This paper presents a 3-way power combined gallium nitride (GaN) MMIC PA using a rectangular waveguide with phase control. The combined PA was fabricated, and its S-parameters and large-signal performance were characterized. To enhance the power-combining efficiency, a novel low-loss dielectric phase controller was implemented to equalize the transmitting phase across the waveguide branches. This dielectric substrate is well-suited for PA fabrication due to its capability to form microstrip lines, facilitating amplifier integration. The power combining technique exhibits low loss, wide bandwidth, and high-power handling capabilities. Measurement results indicate the power-combining efficiency exceeding 73% in the 9-10 GHz band.

Existing two-dimensional Direction-of-Arrival (2-D DOA) estimation methods using Dynamic Metasurface Antenna (DMA) in synchronous mode face challenges due to the limitations of the maximum change rate of DMA states, which can result in DOA estimation failure. In this paper, we propose a novel 2-D DOA estimation scheme, i.e. asynchronous time regulation-based orthogonal matching pursuit (ATR-OMP) for a practical DMA-assisted multi-user, multipath communication system to overcome this issue. The 2-D DOA estimation is formulated as a sparse parameter recovery problem, where the ATR-OMP algorithm is used to extract single-user element data from multi-user beamspace data and estimate the elevation and azimuth angles. Simulation results show that the proposed method has better estimation performance.

Microwave surgical energy devices emit microwaves into biological tissue, causing a temperature rise from within the tissue, which results in tissue coagulation, hemostasis, and vascular anastomosis. The heating region generated by such devices tends to spread outward around the antenna. Because of this characteristic, even when a deeper, vertically distributed heating area is desired, surface expansion of the heating region cannot be avoided, potentially causing damage to surrounding healthy tissue. In order to prevent this, an energy device that minimizes surface energy distribution is needed. In this study, we propose a microwave energy device incorporating a waveguide structure. The waveguide in the proposed device is filled with dielectric material and features a miniaturized cross-section. Furthermore, a ridged waveguide is adopted to generate a localized electric field at the tissue surface. Based on the specific absorption rate (SAR) distributions obtained through both numerical simulations and experiments, itwas confirmed that the microwave energy device using a ridged waveguide can generate a localized heating region effectively.

This paper proposes a new configuration of single-layer dual-band dual-beam reflectarray antenna (RA) that can form two beams to two directions at two frequency bands, depending on the polarization of incident wave from primary horn antenna. To our best knowledge, such dual-beam RAs have not been realized at dual bands yet. To design the element geometries at two bands independently, the proposed RA consists of the two orthogonally arranged dipole elements for a lower band and the Jerusalem cross elements for a higher band. As an example, we present a single-layer dual-beam RA at 15/28-GHz bands, which is constructed by dipoles and Jerusalem cross elements designed to form two beams with a directional difference of 10 degrees at 15 GHz and 5 degrees at 28 GHz. The effectiveness of the proposed RA is proven through the comparison of radiation characteristics between EM-simulated and measured ones.

In this paper, we studied on the signal integrity of a novel 2.3D package platform, called the panel level package (PLP) interposer system. We adopted a simple model which consists of two dimensional transmission line models and fitted parasitic self/mutual inductances and capacitances. Our model is then correlated with the S-parameter model extracted from the real design and the measurement of the fabricated sample. The results show strong correlation in terms of return loss and time domain reflectometry (TDR). The signal integrity of the conventional package based on the Ajinomoto build-up film (ABF) substrate is compared with that of the PLP interposer system. We varied parameters including trace geometry, loss tangent, and parasitic capacitance in our model to improve the signal integrity of the PLP interposer system compared to conventional package. By increasing the cross sectional area of the signal trace and decreasing the parasitic capacitances, the signal integrity of the PLP interposer system can be enhanced to be comparable to the conventional package up to 20GHz, and can be even better in higher frequency ranges.

Through the simulation studies, we propose novel N-channel and P-channel reverse-conducting lateral IGBTs (RC-LIGBT) on bulk silicon with high current drive capability to operate a common-emitter complementary inverter using 200 V power supply instead of the conventional inverter using only N-channel MOSFETs or IGBTs with the high and low sides. The proposed N-channel and P-channel RC-LIGBTs on bulk silicon feature double and triple buried layers respectively, to confine minority carriers within the drift layers and prevent carrier outflow to adjacent circuits or backside, which has been a serious problem for lateral IGBTs on bulk silicon. Moreover, the complementary RC-LIGBTs employ a simple thyristor structure with a P-type or N-type floating base for reverse conduction, which suppresses snapback in forward conduction and reduces the area penalty even in the RC-LIGBT structure. The proposed N-channel and P-channel RC-LIGBTs are applied to a common-emitter type complementary half bridge inverter, and the simulation analysis verifies that the complementary inverter exhibits no penetration current even with a simple gate driving, compared to the conventional inverter which requires complex dead time settings between the high-side and the low-side gate drive circuits.

This paper proposes a boundary integral equation to analyze two-dimensional metallic hollow waveguides. The proposed integral equation has a similar form to that for electromagnetic scattering problems although the unknown function is modified. The infinitely long boundary of the proposed integral equation can be truncated to finite length when the boundary element method is applied to it. The proposed integral equation can be solved combined with the auxiliary equations that are derived by using the orthogonality of modes between the proposed integral equation and a guided or evanescent mode.

In order to validate the proposed method, we perform numerical calculation for three asymmetric waveguides, single step, N-step, and taper waveguides. Numerical results by our method are in good agreement with those by the mode matching method. Moreover, all numerical results satisfy the law of energy conservation to an accuracy of five decimal places.

Many memory-bound AI applications, including natural language processing, transformer-based visual recognition, and multi-task online inference, rely heavily on large-scale general matrix-vector multiplication (GEMV), which is characterized by strong data locality. However, existing hardware architectures for AI model inference face significant data transfer overheads and fail to fully exploit the data locality inherent in these algorithms. We propose a scalable one-logic-two-DRAM (1L2D) multi-core near-DRAM computing accelerator based on 3D hybrid bonding for AI models. Our 3D integration of RISC-V processors with vector accelerators and DRAM presents a unique approach that significantly boosts bandwidth while reducing energy consumption. A memory access circuit supporting page hit mechanism and prefetching strategy is designed to maximize the utilization of the data locality achieved by the algorithm's partitioning and rearranging of data. An interleaving memory address mapping scheme is designed to effectively enhance the bank-level parallelism of data access. Compared with the high-performance Intel Xeon-6230 CPU and the state-of-the-art commercially available UPMEM-PIM, the proposed architecture's computational efficiency for large-scale GEMV is improved by 3.4× and 2.2×, respectively. The architecture achieves a 3.07× improvement in bandwidth and a 76% reduction in energy consumption over the HBM2-PIM.

Brillouin optical correlation-domain reflectometry (BOCDR) enables distributed strain and temperature measurements along optical fibers with high spatial resolution and random-access capability. Conventional BOCDR setups typically require sinusoidal laser frequency modulation, achieved either by direct or external modulation. In external modulation, double-sideband modulators (DSBMs) are generally used, but one of the sidebands must be discarded using narrowband optical filtering. Here, we propose and experimentally validate an approach that replaces the DSBM with a single-sideband modulator, eliminating the need to discard a sideband. We demonstrate efficient single-sideband modulation in BOCDR and apply the technique to measure the temperature distribution in a 100 m silica fiber.

This paper presents a design method for a wideband multimode primary horn by controlling the wavefront on the aperture. This approach minimizes frequency-dependent variations in gain and beamwidth through curved taper optimization. As a design example, we present a spline-profile horn antenna optimized over a frequency range of 9-16 GHz (fractional bandwidth 56 %), achieving rotationally symmetrical radiation patterns with minimal variations in beamwidth and gain.