As the critical solid-state sources in several terahertz applications including imaging，wireless communication and radio astronomy^［1-3］，Schottky diode-based multipliers have been well-developed over the past years. The coming generation of heterodyne array instruments toward frequencies up to 1 THz directly put forward the requirement of higher local oscillator（LO）output power levels especially in critical pumping frequencies to enable related array receivers ^［4］. Generally，terahertz solid-state sources rely on several individual multipliers in cascade to achieve enough frequency multiplication，in which the efficiency and output power of each stage directly affect the overall performances. For example，the generation of powers at frequencies up to 670 GHz put forward the urgent requirements on sufficient powers over 200~230GHz range. Thus，improving the power-handling capabilities as well as conversion efficiencies have been a research focus of diode-based multipliers.

At the moment，the classical balanced doublers^［5，6］proposed by Erickson are preferred for the inherent suppression on odd-order harmonics，whose power handling-capabilities are restricted by the limited diode numbers. As a result，the power-combined techniques become one of the most preferred methods improving the overall power-handling capabilities，at the expense of non-negligible decreases in conversion efficiency ^［7-9］. And the overall efficiency of power-combined doublers would also degrade with the increase of multiplying channels，which restricts the highest achievable output power. Thus，developing novel doubling configurations，which can increase the circuit power-handling capabilities with improved conversion efficiencies，becomes an attractive solution for high power terahertz multiplying sources.

In this letter，a novel four-port balanced doubler prototype presented in our recent work^［10］ has been introduced to break through the achievable output power in G-band（as shown in Fig. 1），which features doubled power-handling capability with improved conversion efficiency than the conventional ones. Accordingly，a hybrid balanced doubler operating up to 230 GHz has been designed and implemented based on improved electro-thermal diode modelling techniques. The proposed circuit could not only inhibit the odd- and fourth-order harmonics，but also can increase the power capacity by a factor of two due to the doubled diode numbers. Namely，the output power could be improved by a factor of two with better conversion efficiency. Moreover，this on-chip concept could drastically reduce its sensitivity to the accuracy of manually assembling，especially at higher terahertz frequencies.

According to the ideal prototype in Fig. 1，the waveguide-based balanced four-port topology could be established，as shown in Fig. 2. This structure is similar but totally different from the circuit in ^［11］，which could provide additional suppression on the adjacent 4^th-order harmonics. The developed circuit employs two pairs of discrete diodes driven in phase quadrature by a hybrid coupler，which are soldered on a 50 μm-thick quartz substrate. Thus，each diode pair in anti-series would be excited under different polarities，resulting in different harmonic currents I₁ （θ）and I₂ （θ）（where θ could be 0° and 90°）. As shown in Fig. 2，the output current of each diode pair should be the differential of I₁ （θ）and I₂ （θ），which inherently suppress the odd-order harmonics. Furthermore，the final current would also be the differential output of（I（0°）and I（90°）），which inhibits the 2N-order（where N is even number）harmonics. The generated harmonic currents are illustrated in Fig. 2，while the related expressions are shown in Fig. 1.

To implement the 220 GHz balanced doubler，four discrete GaAs Schottky diodes have been utilized. As shown in Fig. 3，the utilized diodes feature 300-nm N- GaAs layer and 2.5-μm N+ GaAs layer with a doping concentration of 2*10¹⁷ cm^-3 and 5*10¹⁸ cm^-3，respectively. To increase the overall power capacity，each diode is composed of three anodes（each with a diameter of 7μm）in series featuring a dimension of 230μm×48μm×30μm.

In addition，self-heating and current saturation effects under high operation power must be taken into consideration for better simulation accuracy. A self-consistent electro-thermal model is developed to optimize high efficiency circuits from both the electrical and thermal perspectives. The electrical model of the diode，as well as the self-heating effects characterization，are similar to that in Ref.［12］. Differently，the temperature-dependent junction parameters，including saturation current（I_s），ideality factor（η）and series resistance（R_s），are extracted based on temperature-controlled I-V measurements rather than simply approximating with theoretical expressions^［8］. Utilizing the same test setup in ^［9］，the measured I-V and C-V curve of a 3-anode diode under different temperature levels are illustrated in Fig. 4（a）and（b）.

On the one hand，the constant zero-bias junction capacitance is extracted to be 32fF with C-V curves，featuring low dependency on junction temperatures. On the other hand，the temperature-dependent saturation current（I_s），ideality factor（η）and series resistance（R_s）could be calculated and extracted based on the temperature-controlled I-V curves，utilizing the extraction method discussed in our previous work［9］. And the extracted and fitted temperature-dependent parameters are illustrated in Fig. 5（a）and（b）.

Thermal resistance matrix of the multiplier circuit also need to be extracted to establish the dependency between thermal and electrical domain（shown in Fig. 6（a））. Using the method in Ref.［12］，the 3-order thermal resistance matrix（R_th）could be calculated and fitted（in Fig. 6（b））. In this way，response of temperatures on dissipated power in different anodes could be predicted，as shown in Fig. 6（c）. Furthermore，an accurate self-consistent electro-thermal model could be implemented using the symbolically defined device（SDD）component available in the Advanced Design System（ADS）.

The optimum embedding impedances presented to different harmonics at each junction is determined for maximum conversion efficiency with an input power of 25.5 dBm. With the consideration of self-heating effect，the whole circuit is analyzed and optimized with the cooperation of electro-magnetic and harmonic balance simulations. This process is iterated，modifying the circuit geometry each time，to push the impedances presented to each anode close to their optimum values. Fig. 7（a）illustrated the final geometry of the four-port balanced doubler，which could provide good impedance matching for different harmonics（shown in Fig. 7（b））. The whole circuit comprises an orthogonal coupler，a pair of doubler cells and a three-port E-plane differential probe. The diode chips are assembled on the quartz circuit using the flip-chip technique. And a pair of DC-bias circuits are connected together for better balance. And another circuit has been optimized（Fig. 7（c）），which also provides great impedance matching（Fig. 7（b）），based on the Erickson-style balanced topology with the same diode for better comparison.

The fabricated 220 GHz balanced doubler is shown in Fig. 5. A 105~115 GHz multiplier with an output power tuned to 150~170 mW has been used to pump the traditional circuit in Fig. 7（c）. As shown in Fig. 8（a），a peak power of 44 mW at 222.4 GHz is achieved with related efficiency of 30%，while the peak efficiency is 31.2% at 225 GHz. As the doubler in Fig. 7（a）was designed based on the four-port prototype，it is expected to handle higher pumping power. Thus，a 105~115 GHz source with an output power of 196~340 mW has been utilized. Fig. 8（b）shows the measured output power and efficiency versus operation frequency. The peak power is measured to be 111.27 mW at 225.8 GHz with a corresponding efficiency of 33.7%，while the peak efficiency is 39.5% at 218 GHz with a bias of 12 V. This measurement shows little variations compared with the simulation results，proving the validity of the modified electro-thermal model. The lower driven power levels in the lower band cause the decrease in output power and increase in efficiency，as shown in Fig. 8（b）.

It is difficult to evaluate the efficiencies of these two doublers under different operation powers. Thus，a comparison of the efficiencies as a function of dissipated power in each diode（DPID）need to be conducted. In Fig. 9（a），the simulated results indicate that the four-port doubler features improved conversion efficiencies and output powers than the Erickson-style one at the same DPID，with similar junction impedance matching levels achieved in Fig. 7（b）. Also，the measured results in Fig. 9（a）shows that the four-port doubler exhibits obviously higher efficiency than the Erickson-style one at the same DPID level，which directly proves the great improvement in conversion efficiency. Meanwhile，results in Fig. 9（a）and（b）could prove that this four-port doubler features more than twice the output power of the Erickson-style circuit，due to its doubled diode number and improved efficiency. Saturation power measurement is prevented by the limited driven powers.

Table 1 summarize the reported solid-state frequency doublers in adjacent frequencies. Despite the doubled anode number，the doubler developed in this work could deliver higher efficiency compared with the Erickson-style ones^［13-15］. Meanwhile，this four-port frequency topology would provide higher output power when applied with the GaN-based diodes in the future，which represents a valuable method for high power sources.

In summary，the proposed novel four-port doubler has been developed，featuring a peak efficiency of 39.5% at 218 GHz in continuous waves. And the proved peak output power is 111.27 mW at 225.8 GHz，which reaches the highest diode-based multiplying power reported in this range to the author’s knowledge. This topology also exhibits improved efficiency than the conventional one with doubled power-capability. Thus，this configuration holds great potential in developing high power sources for modern radar systems and other terahertz applications.