domingo, 21 de marzo de 2010

HEMT Millimeter-wave Monolithic IC Technology for 76-GHz Automotive Radar

HEMT Millimeter-wave Monolithic IC
Technology for 76-GHz Automotive Radar

This paper describes monolithic IC technologies for developing practical automotive radar systems, covering the HEMT device structure, IC fabrication process, flip-chip assembling, and circuit design. InGaP/InGaAs HEMTs with a 0.15-μm gate were used in millimeter-wave monolithic ICs for the W-band, providing a maximum stable gain of 9 dB at 76 GHz. Height-controlled flip-chip bonding with pillar interconnection was demonstrated as a low-cost assembly method. A de-embedding technique to model a face-down co-planar waveguide was proposed. A chip set was designed with this technique, consisting of a 76-GHz amplifier, 76-GHz mixer, 76-GHz SPDT switches, 38/76- GHz doubler, 38-GHz voltage-controlled oscillator, and 38-GHz buffer amplifier. The fabricated chip set showed high performance for automotive radar systems.

1. Introduction

In line with recent advances in millimeterwave systems based on semiconductor technology development, the application of these systems is being extended to commercial fields.1),2) One of the most promising areas is automotive radar to provide active safety for driving. Research on automotive radar started in the early 1970s in the United States. In 1981, Fujitsu developed a 50- GHz radar sensor. These efforts, however, did not meet the demands of the market due to cost and size. In the late 1980s, several national programs promoted research on millimeter-wave systems, while during the same period GaAs semiconductor manufacturing grew strongly due to the rapid spread of wireless communications. These circumstances stimulated commercial interest in millimeter- wave systems, including automotive radar. Frequency allocation is a key to the market of any system utilizing electromagnetic waves. The Japanese Ministry of Posts and Telecommunications (MPT) assigned the 60-GHz band for low-power millimeter radar systems in 1995. Following the allocation of the 76-GHz band for radar in the U.S. and Europe, the MPT decided in 1997 to assign the same band to automotive radar systems in addition to the 60-GHz band. Although this action promoted a global standard for commercial use, it gave a rise to challenging technical issues. To realize practical, reasonably priced radar systems for passenger cars, the frequency performance of the devices must be improved and costs reduced. Monolithic integration is the usual approach to cost reduction and is still important for the W-band.3) Much effort has recently been devoted to the development of monolithic integrated circuits, and it has been shown that peripheral elements play a more important role in this frequency region. Die positioning and wiring in a package seriously affect the circuit performance and yield, resulting in higher costs. Flip-chips can be used to avoid these difficulties, as well as digital circuits.4) Using flip-chip bonding, chip positioning can be automated with greater accuracy. The wafer process for flip-chips will be simpler than that for conventional microwave devices. To make use of these advantages in millimeter-wave applications, problems associated with the short wavelength must be overcome.5),6) This paper describes a millimeter-wave monolithic IC (MMIC) technology for the development of 76-GHz automotive radar, including a 0.15-μm HEMT IC process, flip-chip with pillar interconnection, and W-band face-down circuit design. Using an InGaP/InGaAs/GaAs heterostructure with electron beam lithography, the HEMT exhibited good improvement compared to the conventional structure. The pillar interconnection technique was employed in the millimeter-wave monolithic IC for flip-chips. We have demonstrated the good electrical properties of the flip-chip and showed the difficulty of design due to the facedown structure. Several MMICs were successfully designed and fabricated for the 76-GHz automotive radar sensor. These face-down circuits exhibited good performance in the W-band.

2. Device technology

The operation frequency of 76 GHz in MMICs is challenging for transistor technology. The transistors in the circuit must provide sufficient gain at 76 GHz, which conventional semiconductor circuits have not reached. Heterojunction devices such as HEMTs have fundamental advantages over other types of transistors in terms of highfrequency performance, because of the bandgap difference in their structure. Fujitsu has already developed a 0.25-μm gate HEMT technology using photolithography and sidewall techniques, which is commercially available for several millimeter- wave systems. However, even with quarter- micron HEMTs, the frequency performance is not sufficient for 76-GHz automotive radar systems; high-performance transistors with an fmax exceeding 150 GHz are required. The reduction of gate capacitance and increase in transconductance are important to improve transistor characteristics. Scaling is the traditional approach for FET technology and is effective for improving performance even in the sub-micron region, but parasitic factors cannot be neglected for high-performance devices. Such phenomena as the short channel effect, increase in gate resistance and fringing capacitance degrade the transistor performance. Figure 1 shows the new structure employed for the 0.15-μm HEMT. The T-shaped gate was patterned by electron beam lithography. This makes the gate electrode free from the sidewall, thus reducing the fringing capacitance. The i- GaAs/i-InGaAs/n-InGaP/n-GaAs epitaxial layers were grown by MOCVD, where n-InGaP is the electron supply layer instead of the AlGaAs layer in the conventional structure. Thanks to high doping in the InGaP layer, the distance between the channel and gate can be reduced compared with the conventional AlGaAs/GaAs HEMT. This structure thus suppresses the short channel effect for fine gate lengths. In the fabrication process, the recess region was etched after ohmic electrode formation. The Al gate electrode was aligned in the recess region with electron beam lithography and lifted off. Using Al for the T-shaped gate electrode, the gate resistance is low even for a gate length of 0.15 μm. We optimized the structure taking account of the high frequency and process stability. Figure 2 shows the small-signal frequency characteristics of a 0.15-μm HEMT with a bias condition of Vds = 3 V and Vgs = -0.4 V, indicating a cut-off frequency (fT) of 90 GHz, a maximum oscillation frequency (fmax) of 170 GHz, and a maxi

mum stable gain (MSG) of 9 dB at 76 GHz. Here, fmax was determined from the maximum available gain curve instead of by -6 dB extrapolation of the unilateral power gain curve. This result shows that the new HEMT structure effectively reduces the parasitic effects. Using the 0.15-μm gate HEMT, we developed an MMIC process for low-cost millimeter wave applications. Similar to the conventional MMICs, MIM capacitors, epitaxial resistors, and air bridge wiring are employed for passive elements. The conventional MMIC uses micro-strips as transmission lines, which need a ground plane on the backside of the chip and via holes to connect the backside ground plane to the face-side of the chip. The thickness of the chip must be precisely controlled to maintain the impedance of the transmission line. This requires a wafer thinning process and wafer penetrating process for the fabrication. The controllability of the process to determine wafer thickness and via size poses a difficult problem to produce much higher frequency circuits with the micro-strip structure. A co-planar waveguide (CPW) is often used in millimeter-wave systems and its structure is also convenient for flip-chip

assembling, since all electrodes in the circuit are placed only on one side of the chip. No wafer thinning or via penetrating process is needed for coplanar circuits. The chip can also be thick, thus retaining mechanical strength during the chip bonding process. Digital LSIs usually use a bump technology based on transferred solder for the interconnections of flip-chips. The MMIC for the W-band requires more precise positioning and size of interconnection electrodes to avoid degradation due to impedance mismatching than digital LSIs because of its short wavelength. In order to obtain good positioning with a small diameter interconnection, we developed a pillar technique that is a part of the fabrication process. After the metallizing process for the MMIC co-planar lines, pillar-shaped electrodes with 40-μm diameter gold are formed on the wafer by electroplating. Figure 3 shows an SEM photograph of the MMIC, showing the

pillars, HEMTs, and an air-bridge on the circuit. This technique provides flexibility, allowing the pillars to be placed anywhere on the chip and thus reduces the source inductance of each transistor by placing a grounded connection just beside the devices. In the fabrication process, pillar plating is also needed, instead of wafer thinning, backside alignment, via hole etching, and back-side plating, which are conventionally used in MMICs. The entire process is thus simplified with this pillar technique.
3. Flip-chip technology for millimeterwave devices
The importance of assembling is much larger for high-frequency applications than for digital or low-frequency ones. The physical length of packages seriously affects the performance of MMICs because the wavelength is comparable to the dimensions of the package and leads. Flipchip bonding causes little loss in performance or size reduction, and its assembly precision enables low-cost fabrication.7)-9) However, two problems remain. Firstly the transmission line on the chip to the substrate exerts a proximity effect on the highfrequency characteristics of the device. Secondly, the reliability of the lines may be compromised when subjected to thermal and mechanical stresses inherent in the inflexible structure of flip-chip bonding. We investigated the high-frequency characteristics and reliability of the flip-chip bonding portion between a GaAs chip and alumina substrate. The gold-tin eutectic reaction method was employed for flip-chip bonding, in which the chip and substrate were heated after connecting the gold pillars on the chip to the tin of the substrate. This process potentially causes less damage to the fragile GaAs chips than conventional gold-to-gold thermocompression bonding7),8) because of its lower temperature and lower load. In addition, the gold-tin method avoids solder contamination of the gold transmission line on the chip, which can be a problem in the solder-bump bonding method.9) The test piece consisted of a 2-mm-square GaAs chip with gold CPW lines and gold pillars on one surface, and a 10-mm-square 0.65-mmthick alumina substrate with a patterned thin film metallization on its surface. Figure 4 shows a schematic diagram of our flip-chip. The substrate metallization consists of 100 nm of Ti, 3,000 nm of Cu, and 500 nm of Sn. The chip was mounted on the substrate with a pulse-heat tool at 350°C for 10 seconds under a 20 g/pillar load.
Figure 5 shows the scattering parameters of the sample: S21 at 76 GHz measures -1.8 dB. By removing the return loss (-0.3 dB), the loss from the substrate coplanar lines (-0.5 dB) and the transmission line on the chip (-0.6 dB), the insertion loss of the pillar portion is estimated to be -0.2 dB/pillar. S11 is below -15 dB at 76 GHz. Using simulation, typical values for conventional wire bonding were estimated at -0.6 dB for S21, and -10 dB for S11.3) These results show that flipchip bonding has both a lower insertion loss and a return loss than conventional wire bonding. Next, we estimated the proximity effect on the high-frequency characteristics of the device by simulating the electromagnetic field using the moment method. Figure 6 shows the simulated characteristic impedance of the transmission line on the chip as a function of the pillar height. The transmission line on the chip has a coplanar structure for a 20-μm gap and 20-μm line width. The characteristic impedance gradually decreases with a decrease in pillar height. A ± 5-μm change in the height of a 20-μm-high pillar results in a ± 2- ohm change in the 52-ohm characteristic impedance. The slope of the curve becomes remarkable below a 15-μm pillar height. This is possibly explained by the change in the transmission mode. The result indicates that the flip-chip needs bonding technology with a controlled air gap, especially at a low pillar height such as 10 μm. The controllability closely depends on the load in the bonding process. Gold-tin eutectic bonding is suitable because of its low load. The pillar adherence strength was determined by measuring the die shear strength of the flip-chipped GaAs chips. The pillar adherence strength was equivalent to or larger than the TAB bump adherence strength.7) In general, the major problem in flip-chip bonding is the thermal expansion mismatch between the chip and substrate, because both are inflexible and thermal stress concentrates at the bonding portion. GaAs has a thermal expansion coefficient (6.0 ppm/°C) similar to that of alumina (6.5 ppm/°C), which is a typical substrate material. The flip-chip bonding of GaAs devices on an alumina substrate is expected to have a relatively small thermal expansion mismatch. The bonding reliability was estimated by thermal cycle tests (-55 to 150°C; 1,000 cycles), thermal shock tests

(0 to 100°C; 1,000 cycles) and vibration tests (10 to 200 Hz; 10 G, 6 hours). All samples passed these tests, indicating that the bonding technology has a high reliability. Figure 7 shows a cross-section of the flip-chip bonding after the thermal cycle test.
4. Circuit Design
4.1 Face-down chip design
Designing the W-band millimeter-wave circuit with face-down chips is another challenge. A precise model of the transmission line is one of the key issues for MMICs. In spite of much research on CPW design,10),11) there remain several difficulties in using CPW for flip-chip circuits, such as the proximity effect and signal transition to the chip. An actual flip-chip CPW line contains signal transition parts such as pillars or bumps to transmit a signal from the external feed line to the CPW line. When characterizing the actual CPW lines, however, the signal transition part hinders the de-embedding of the external feed lines since the transverse electromagnetic field extends toward each line. Figures 8(a) and (b) show a portion of an actual flip-chip structure, which includes a signal transition pillar. A simulated result of a magnitude contour map shows the electric field distribution at a distance of 10 μm from the alumina substrate at 76 GHz. The simulation was performed using a 3-dimensional full-wave electromagnetic field simulator. As shown in Figure 8(b), the electric field expands under the CPW line on the GaAs chip. This leads to interaction between the signal transition pillar and CPW line, thus making it difficult to determine the reference plane when de-embedding each line. When designing the circuits, the pillar transitions and CPW lines must be characterized independently. Therefore, we propose an experimental technique to model flip-chip CPW lines without signal transition pillars. This experiment is based on the reversed-structure flip-chip, which consists of a CPW line chip and lid ground. The CPW lines are measured with the lid ground attached to the pillar. By using this technique, the characteristics of the transmission line can be evaluated without affecting signal transition through the pillars. We obtained that Z0 = 49.0 ohms, Eeff = 5.41, and Attn= 0.300 dB/mm from measured data for a line width of 20 μm, a ground-to-ground spacing of 60 μm, and a pillar height of 20 μm. We also obtained that Z0 = 55.9 ohms, Eeff = 6.33, and Attn= 0.241 dB/ mm for the same line width and ground-to-ground spacing for a line without the lid ground. As mentioned in the previous section, the CPW line and ground on the substrate interact. These results closely agree with the simulation results and indicate that W-band flip-chip MMICs with CPW lines can be designed using the quasi-TEM mode approximation. Thus, we identified the parameters for the transmission line model in the circuit simulator by using this method.

4.2 MMIC chip set for automotive radar
Based on the CPW model, we developed a 76- GHz chip set for automotive radar. The chip set consists of a 76-GHz amplifier, 76-GHz mixer, 76- GHz SPDT switches, 38/76-GHz doubler, 38-GHz voltage-controlled oscillator, and 38-GHz buffer amplifier. Figure 9 shows photographs of the chip set. The amplifier has a two-stage configuration

with 80-μm HEMTs. Short-circuited stubs are used in the inter-stage and output matching circuits to reduce the chip size. Each short-circuited stub is composed of a transmission line and a grounded MIM capacitor. Figure 10 shows the measured S-parameters of the amplifier. The amplifier has a small-signal gain of 10.6 dB at 76.5 GHz

The mixer has a singly balanced configuration. The RF and LO signals are fed into the gate of each HEMT through the 0-p hybrid circuit. The 0-p hybrid circuit consists of a quarter-wavelength transmission line and a branch line hybrid circuit. The mixer using the 0-p hybrid circuit exhibits good isolation between the RF and LO ports even if the unit mixers have large reflection coefficients. Also, the mixer reduces AM noise from the LO in cooperation with an external IF 180- degree hybrid circuit. The mixer has a conversion gain of -4 dB at a 76-GHz LO frequency for an LO power of -1 dBm. The 76-GHz SPDT switch consists of two parallel- resonated 160-μm HEMTs. The transmission line connected to the drain and source terminals of the HEMT is employed as a resonant inductor in parallel with the fringing capacitance. The obtained insertion loss and isolation at 76.5 GHz were 3.4 dB and -17 dB, respectively. The 38/76-GHz doubler chip consists of two single-ended 80-μm HEMT frequency doublers and a branch line hybrid circuit. This circuit con- figuration leads to a low input return and high isolation between the two output ports. Figure 11 shows the measured frequency response of the doubler. The doubler has a conversion gain of -3.7 dB for a 38.25-GHz input signal power of 3 dBm. In the 38-GHz voltage controlled oscillator, we adopted the reverse channel HEMT configuration to generate sufficient negative resistance and reduce the size of the source feedback circuit. The oscillator uses the transmission line resonator terminated by a 160-μm Schottky-barrier diode as a varactor. The 38-GHz buffer amplifier provides impedance isolation between the oscillator and the external load. The chip has an output power of -0.7 dBm at 39.6 GHz. Figure 12 shows an SEM photo of the flipchip MMIC, indicating the amplifier chip on the evaluation board.

5. Conclusion
We have successfully developed an HEMT MMIC technology for 76-GHz automotive radar. The wafer process is based on a 0.15-μm InGaP/ InGaAs HEMT and interconnecting pillars for flipchip bonding. The maximum stable gain of the HEMT is 9 dB at 76 GHz. The flip-chip bonding technology uses gold micro pillars and tin metallization on an alumina substrate. Results of reliability tests indicated that the flip-chip bonding is very resistant to thermal and mechanical stresses, and thus is suitable for practical use. For designing flip-chip mounted MMICs for the W-band, we propose an experimental method to model facedown CPW transmission lines. By using this model, the chip set was designed and fabricated, and achieved sufficient performance for 76-GHz automotive radar systems. The results indicated that our MMIC technology is promising for achieving low-cost 76-GHz automotive radar systems.
Luiggi Marquez C.I 17677911 CRF

¿Sabes que la Videollamada de Messenger es GRATIS ¡Descúbrela!


A range of low-noise amplifiers have been designed for both the new and the proposed observing bands for the Australia Telescope National Facility Compact Array antennas. The amplifiers, which operate in three bands in the range 16 GHz to 115 GHz, use indium phosphide based high electron mobility transistor, monolithic microwave integrated circuit technology. We will describe the amplifier design and report the room temperature on-wafer performance of these amplifiers.

1 Introduction

The Australia Telescope Compact Array, situated near Narrabri in northern New South Wales, consists of six 22-metre dish antennas. Five of the antennas are moveable on an east-west rail track three kilometres in length, while the sixth dish is fixed at a position three kilometres further to the west. At present, the Compact Array can operate at a range of frequencies in the 1 to 10 GHz band. The Australia Telescope National Facility is currently enhancing the frequency coverage of its Compact Array. Receiver systems covering two new observing bands, the 12 mm band (16-25 GHz) and the 3 mm band (85-115 GHz), are being designed and will be installed in the ATCA antennas. Provision has been made to add, later, a 7 mm band (30-50 GHz) to the new receiver package. A number of low-noise, indium phosphide based, monolithic microwave integrated circuit (MMIC) amplifiers have been designed to cover the new observing bands. Indium phosphide based high electron mobility transistors (InP HEMT), cooled to cryogenic temperatures, have been used in the construction of wide band, millimetre-wave receivers with noise performance comparable with SIS mixer receivers. Cryogenically cooled InP HEMT MMIC lownoise amplifiers have been reported with noise figures as low as 0.4 dB at 100 GHz. 2 Amplifier design, fabrication and performance The MMIC amplifiers were designed for the advanced, 0.1 micron, InP HEMT process recently developed at TRW. As the lowest noise performance has been reported for cryogenically cooled amplifiers, we intend to cool these amplifiers, but the transistor models are only valid for a range of temperatures around 25°C, and do not explicitly model device performance at cryogenic temperatures. Consequently, circuit performance was optimized for low noise, flat gain, good input and output return loss and stability using the (room temperature) models that were available. When indium phosphide based transistors are cooled, their transconductance increases, leading to greater gain and lower noise. Some circuits, which are quite stable at room temperature, may become unstable, due to the increase in device gain, at cryogenic temperatures. This was a concern when the amplifiers were designed, as it is difficult to assess the stability of a circuit at cryogenic temperatures, especially at the higher frequencies where active device models and passive element models may not be sufficiently accurate. As a result, a conservative approach was taken in the design, with every effort made to ensure the stability of the resulting circuits at cryogenic temperatures. In general, the stability factor, K, was kept greater than 4 at frequencies within the nominal band of the amplifier, and greater than 10 outside that band. Source inductance was included in all HEMT stages to aid in levelling the gain across the band and to improve the amplifier stability. The source inductance in the first transistor is generally greater than that in the other transistors, and allows the input return loss of the amplifier to be increased while maintaining low noise figure in the design. Relatively large capacitors (10 pF in parallel with 1.2 pF) are used for RF bypass in the gate and drain bias networks. Small series resistances are used in the gate and drain bias lines to damp out any parasitic resonances. The gain, input return loss and output return loss of the amplifiers, reported below, have been measured on-wafer, and only at room temperature.

2.1 The 16-25 GHz amplifier This amplifier was designed for minimum noise in the 16 to 25 GHz band with flat, 30 dB gain and input and output return losses greater than 15 dB over the whole band, and better  han 20 dB midband. The circuit uses three HEMTs, each with a total gate width of 120 microns, and is 3.1 mm x 2.25 mm. The simulated performance of the amplifier is shown in Fig. 1(a). Fig. 1(b) shows the layout of the amplifier circuit. Typical amplifier performance, shown in Fig. 1(c), is remarkably similar to the simulation. The main difference is that the measured input return loss near 25 GHz is 10 dB, rather than 18 dB as simulated. The noise figure of the amplifier, measured on-wafer up to 25 GHz, is

shown in Fig. 1(d). The noise figure is somewhat higher than that predicted in the simulation. 2.2 The 30-50 GHz amplifier This amplifier was designed for minimum noise in the 30 to 50 GHz band with flat, 30 dB gain and input and output return losses greater than 15 dB over the whole band, and better than 20 dB midband. The circuit uses a 4-finger HEMT, with a total gate width of 120 microns, in the first stage, followed by 4-finger HEMTs, each with a total gate width of 80 microns, in the three subsequent stages, and is 2.7 mm x 2.25 mm. The simulated performance of the amplifier is shown in Fig. 2(a). Fig. 2(b) shows the layout of the amplifier circuit. The typical amplifier performance, shown in Fig. 2(c), is, again, similar to the simulation. The measured input and output return losses are both lower than that predicted by the simulation, especially at the high frequency end of the band. The noise figure of the amplifier, measured on-wafer between 30 and 39 GHz, is shown in Fig. 2(d).

2.3 The 85-110 GHz amplifier This amplifier was designed for minimum noise in the 85 to 115 GHz band with 12 to 14 dB gain up to 110 GHz. Input and output return losses were designed to be greater than 10 dB in the frequency range 92 to 115 GHz, and better than 15 dB in the frequency range 95 to 105 GHz. The circuit uses four, 4-finger, HEMTs, each with a total gate width of 40 microns, and is 2.1 mm x 2.25 mm. The simulated performance of the amplifier is shown in Fig. 3(a). Fig. 3(b) shows the layout of the amplifier circuit. The typical amplifier performance is shown in Fig. 3(c). The measured input and output return losses are both lower than that predicted by the simulation. The gain is higher than predicted, but falls off quickly at the high frequency end of the band. The noise figure of the amplifier, between 90 and 98 GHz, is shown in Fig. 3(d). The estimated accuracy of the on-wafer noise figure measurement is ±1 dB. The noise figure is similar to that predicted by the simulation.

Low noise amplifiers, operating in three bands in the range 16 GHz to 115 GHz, have been described. The accuracy of the simulations indicates that circuits can be designed to work up to 120 GHz by relying on standard component models and foundry device models, with higher simulation accuracy being achieved at the lower frequencies.

Luiggi Orlando Marquez C.I 17677911 CRF

¿Quieres saber qué móvil eres? ¡Descúbrelo aquí!

Wideband Monolithic Microwave Integrated Circuit Frequency

Wideband Monolithic Microwave Integrated Circuit Frequency
Converters with GaAs mHEMT Technology

Abstract—We present monolithic microwave integrated circuit (MMIC) frequency converter, which can be used for up and down conversion, due to the large RF and IF port bandwidth. The MMIC converters are based on commercially available GaAs mHEMT technology and are comprised of a Gilbert mixer cell core, baluns and combiners. Single ended and balanced configurations DC and AC coupled have been investigated. The instantaneous 3 dB bandwidth at both the RF and the IF port of the frequency converters is ∼ 20GHz with excellent amplitude and phase linearity. The predicted conversion gain is around 10 dB. Simulated results are supported by experimental characterization. Good agreement is found between simulations and experiment is found after adjustment of technology parameters.


Ultra-wideband MMICs based on mHEMT technology operating over an instantaneous bandwidth of more 20 GHz are discussed in this paper. The availability of such MMICs reduces front-end complexity and enables reuse of components for different applications in receivers for radar, radiometer and communication applications. One of the key elements in heterodyne receivers is the frequency converter circuit. Double balanced structures based on Gilbert cell mixers have been used earlier and wideband operation has been demonstrated by the authors in SiGe and GaAs technology. Several highly integrated active mixers based on FET technologies have previously been reported. These active mixers have mostly been optimized for down-conversion performance, whereas the active mixers demonstrated here can be used for both upand down-conversion. This paper focuses on fully integrated DC and AC coupled frequency converters comprised of a Gilbert cell active mixer, balun circuits, balanced buffer amplifiers, and combiners. Aspects of mixer design, such as balance considerations have been discussed in. In this paper we will discuss transistor scaling issues and stability problems. Stability becomes especially important when moving to faster technologies such as e.g. the GaAs mHEMT D01MH process from OMMIC employed here. We also compare the predicted performance versus experimental results.


The aim of the DC coupled frequency converter was to demonstrate ultra-compact wideband MMIC operation reaching the mm-wave range with competitive performance. In contrast to other circuits, this layout utilizes efficiently the wafer space and is very compact. Fig. 1 shows the layout of the circuit including a Gilbert cell active mixer, two baluns for the LO and RF port, respectively, and a buffer amplifier with source followers and level shifters. All stages are DC coupled and the DC level is adjusted for optimum operation. The circuit is fully differential on input and output. The input matching is achieved by additional 50 Ω resistance together with a short transmission line and buffer amplifiers.

The basic circuit for the fully differential frequency mixer is shown in fig.2 together with the buffer circuitry. The RF and LO ports are buffered and can be accessed either differentially or single ended. Especially for the LO port this is convenient if a quadrature oscillator is employed with differential output, which is frequently the case in integrated down/up converters. The output IF port is also differential and is buffered through a buffer amplifier. The buffer amplifier is illustrated in fig 3.

A similar circuit has been used for the baluns, which are integrated on another chip and which occupy roughly quarter of the space of the mixer.

The predicted conversion gain performance versus frequency can be depicted from fig. 4 together with the insertion loss at the RF port. The LO and RF ports are almost identical and well matched up to mm-wave frequencies. Simulations predict a 0.2 dB bandwidth of 12 GHz. However, it can be clearly seen that a resonance exist at around 37 GHz. This resonance gives rise to instability problems. Its origin is an interconnect line between the mixer output and the source followers in the buffer circuit. The line length is close to λ/4 at this frequency. Eliminating the contribution of the line by reducing

the line physical length or change in transmission line impedance, the circuit is absolutely stable over the whole frequency range. The simulated resonance can also be detected in measurements, however due to the losses and a slight shift in the processing parameters the resonance does not give rise to oscillations in measurements. Measurements have been performed with regards to frequency conversion parameters. The RF port conversion bandwidth was measured by sweeping both the RF and LO signals at a constant IF frequency of 1.2 GHz. The IF bandwidth has been measured by keeping the LO signal at a constant frequency and changing the RF signal frequency. The results shown in fig.5 have been corrected for the cable losses in the measurement system. Both ports exhibit a relatively flat characteristic as a function of frequency. Again the circuit works nonoptimally due to non-optimum performance of the current sources. Nevertheless, the operating 3 dB bandwidth for both ports is close to 15 GHz even in this non-optimum case.

One of the major issues in achieving the good results is the proper scaling of the transistors for the transconductance and the switching stage. This point will be discussed in detail in the full paper.
The AC coupled MMIC frequency converter is a modified version of the circuit presented in. It utilizes active baluns based on a common-gate common-source circuit. Input matching in the circuit is achieved with the common-gate FET of the active balun. The GaAs MMIC exhibit slightly wider bandwidth as compared to the SiGe HBT technology. They also exhibit a larger dynamic range. The output of the mixer is DC coupled and inductive peaking is employed for bandwidth enhancement, which is important due to the capacitive loading of the mixer output. The combiner is a modified differential amplifier with wideband performance, level shifter, and source follower stages. The conversion gain versus frequency is shown
in fig. 6. The predicted conversion gain is around 10 dB and is expected to be flat up to frequencies of 20 GHz. The measured values are significantly different. First, the measured conversion gain was determined to be around -3 dB. Secondly, a strong frequency dependence is visible. Detailed investigation revealed that the common gate mHEMT of the balun exhibited a problem. Taking this into account and including a threshold voltage shift from the nominal VT = −1.0V to VT = −1.15V gave excellent agreement between measured and simulated values. This is verified by the DC bias current characteristic versus the supply voltage as indicated
in fig.7

An interesting feature of this circuit is among others that the noise properties of the converter can be optimized by appropriate choice of the common-gate transistor periphery and bias conditions. Therefore, low-noise operation is feasible with such a circuit However, the linearity is still a concern and will be further investigated in the future.
MMIC circuits with ultra-wideband operation have been designed and fabricated. Results for down/up-converters show a flat conversion gain with a 3 dB bandwidth in excess of 20 GHz. The converter circuits exhibit similar bandwidth at all ports, including the IF port. Therefore, one can employ these circuits as up/down converters with minimum redesign. Very compact circuits have been successfully characterized. Good agreement between simulation and measurement is obtained if the parameters are adjusted for technological spread. Problems with some transistors have been observed in the design, as well as stability problems, due to the large mismatch between the mixer output stage and the source followers. The origin of the instability has been identified and potential stability problems can be alleviated by reducing the interconnect line length between the output of the mixer and the source followers of the buffer amplifier. The presented circuits demonstrate the feasibility of a compact fully integrated receiver front-end in GaAs technology spanning the frequency range from DC to Ka band. The interesting aspect of this work that the same chip can be used for up-conversion as well as down-conversion.

Luiggi Marquez C.I 17677911 CRF

¿Quieres saber qué móvil eres? ¡Descúbrelo aquí!

Diamond Heat-Spreaders for Submillimeter-Wave GaAs Schottky Diode Frequency Multipliers

Diamond Heat-Spreaders for Submillimeter-
Wave GaAs Schottky Diode Frequency

Abstract—We have attached CVD diamond film as a heatspreader to our existing 250 GHz GaAs Schottky diode frequency tripler chip in o rder to improve its power handling capability. Using this first generation device, we were able to produce 40 mW at 240 GHz from a single frequency tripler with 350 mW input power at room temperature. We have also run finite-element thermal simulations and seen that the 30 μm thick diamond film dropped the temperature of the anodes by about 200° C. This break-through in thermal management increases the output power of frequency multipliers by nearly 100%.


The planar GaAs Schottky diode frequency multiplier chain is a very good candidate for compact reliable tunable terahertz sources due to low mass, wide electronic tunability, narrow line-width, low noise, and room temperature operation. However, the output power of frequency multiplier chains is relatively low compared to recent quantum cascade laser (QCL) results in the 2-3 THz band. Frequency multipliers are cascaded in chains to create terahertz sources. Thus, in order to generate high power at the output in the 2- 3 THz range, we need to have high driving power in the first stage of the chain, typically in the 200 to 400 GHz range. In order to increase the power handling capability and output power, the doping concentration of the Schottky contact can be adjusted to increase the breakdown voltage, or the number of anodes per multiplier chip can be increased. However, due to thermal issues the most promising technique to increase output power has been to power-combine two or more parallel stages. For example, the power available at 300 GHz can be doubled by using a 300 GHz dual-chip power combiner. Currently, up to 800 mW of continuous-wave (CW) tunable W-band power is available by four-way power combining MMIC amplifier chips at W-band. However, currentgeneration submillimeter-wave frequency multipliers cannot handle the input power of 800 mW due to thermal management problems. Therefore, we propose to use a diamond film as a heat-spreader in order to remove the heat more efficiently, resulting in increased power handling and output power.

Figure 1 (a) shows the nominal schematic view of the 250 GHz tripler as placed in the split waveguide block. The heat generated at the anode transfers through a few-micrometer thick GaAs membrane and gold beamlead to the waveguide block, which is a heat sink. The thermal conductivity of GaAs is approximately 46 W/m•K and this thermal conductivity value decreases as the temperature increases. Figure 1 (b) shows the proposed schematic view of the 250 GHz tripler with a diamond film bonded to the backside of the GaAs substrate to remove the heat more efficiently. The polycrystalline diamond film works as a heat spreader which removes the heat by thermal conduction to a heat sink. In this structure, the heat goes to the diamond film and then transfers through the diamond film laterally and goes back to the gold beamlead and waveguide block which is a heat sink. The thermal conductivity of CVD diamond is 1000-1200 W/m•K and is 20 times greater than GaAs (46 W/m•K) and three times higher than silver (430 W/m•K). Since the thermal resistance of the chip is about three times lower with diamond than without, it reduces the maximum temperature rise by a similar factor.


A. Diamond Etching Process

Polycrystalline diamond deposited by hot-filament chemical vapor deposition (CVD) has been selected as the material for providing thermal management due to the high thermal conductivity (1000-1200 W/m•K). It is an electrical insulator (resistivity = 1015 W/cm) and has a moderate relative dielectric constant of 5.7. Because it is the hardest material known and it is chemically inert, it is extremely difficult to pattern the diamond, especially thick films. In order to etch the thick diamond film, the RF power, the gas flow, chamber pressure, and the bias voltage have been investigated in inductively coupled plasma (ICP) RIE. An etch rate of 550 nm/min has been used.

B. Microfabrication

JPL's Monolithic Membrane-Diode (MOMeD) process that results in extremely low parasitic Schottky diode chips has been discussed previously. The frontside processing forms the Schottky diode, RF components and on-chip capacitors. The backside processing is used to remove the substrate and enable chips that are made on a very thin layer of GaAs to improve RF tuning. We have modified the backside processing sequence to allow us to mount a diamond film to the membrane. The diamond is patterned using an Inductively Coupled Plasma Reactive Ion Etching (ICP-RIE) system. The patterning of the diamond allows us to shape the diamond substrate, keeping the beamleads free for mounting and DC contacts. Figure 2 shows the front-side view of the 250 GHz tripler. The front-side view is the same as the tripler without the diamond. However, the back-side has the bonded diamond film for removing the heat efficiently.


Figure 3 shows the measured 250 GHz tripler's output power and conversion efficiency as a function of the input power at 238.8 GHz. At 200 mW input power, the efficiency

was approximately 14 % for one with the diamond, and 11% for one without diamond. However, above 200 mW of the input power, the efficiency drops more slowly for the tripler with the diamond compared to the tripler without the diamond. The output power of the tripler without diamond peaks at 22 mW for the input power of 200 mW and thereafter starts to drop rapidly and fails at the input power of 240 mW. However, for the tripler with the diamond heat-spreader, the output power continues to climb as the input power increases. An output power of 40 mW at 350 mW input has been achieved from a single chip without any sign of degradation. In addition, reduction of operation temperature of Schottky diodes can increase the reliability. Figure 4 shows the output power versus output frequency plot of the 250 GHz tripler for a chip with diamond and a chip without diamond. A frequency sweep was performed using flat input power levels at 100 mW, 200 mW, and 300 mW. The tripler without the diamond suffered a catastrophic failure at about 250 mW input power, and therefore no data was obtained for this chip at 300 mW. As the plot shows, adding the diamond layer improves the output power without degrading the bandwidth.


This superior thermal management achieved with diamond provides a 100% increase in power handling capability. For example, we have achieved 40 mW output power at 240 GHz from a frequency tripler with 350 mW input power, while identical triplers without diamond suffered catastrophic failure at 250 mW input power. Optimizing the Schottky diode chips and waveguide circuits for the presence of the diamond substrate is expected to further increase the achievable output power. This increase in output power in the 250-350 GHz band is expected to increase the usable range of Schottky diode frequency multiplier chains to beyond 3 THz

Luiggi Marquez C.I 17677911 CRF

Compartir tus mejores FOTOS es fácil en Messenger ¡DESCUBRE cómo!