Integration of a K-Band Receiver Front-End Using a Copper Core Printed Circuit Board

In recent years, various applications, for example VideoOn-Demand and cloud computing, have led to research of high bandwidth satellite communications. High gain antenna arrays with beamsteering capabilities allow users on mobile platforms such as cars or trains to enjoy high speed communications provided by satellite data links.

The K/Ka-Band satellite communications bands, with Ka-Band uplink and K-Band downlink, has the potential to offer a large bandwidth to mobile platforms. However, this introduces many challenges regarding the beamsteering capability for terrestrial satellite communications platforms. Electronic beam steering, made of active antenna arrays, is one of the possible options to arrive at highly integrated, robust satellite terminals. In active antenna arrays, each antenna element or group can be controlled in amplitude and phase and, due to scanning requirements, the distance between antenna elements in the array is typically close to λ/2. Hence, for higher frequencies, the available space for the front-end components decreases, while chip size may not decrease. This leads to significant chip integration challenges. The integration requires a sophisticated packaging technology that offers (1) excellent high-frequency properties, (2) high integration density, and (3) a highly conductive thermal path from the chips to a heat-sink and subsequently excellent heat management.

Most examples of K- and Ka-band MMIC integration utilize low temperature co-fired ceramics (LTCC) substrates with wirebond technology. For example, in [1], a single transceiver with a low-noise amplifier (LNA), a power amplifier (PA), switch, phase shifter, and attenuator were packaged in cavities with bondwires in LTCC technology. For [2], Spira integrates four front-ends into an LTCC package for beam steering applications. The integration strategy in [3], also utilizes LTCC. Holzwarth et al. have also proposed a beamsteerable antenna array in LTCC, which addresses significant cooling challenges with a complex water cooling integrated into the carrier. This solution implements digital beam forming, which introduces significant processing complexity in the back-end [4]. In [5], an LTCC package was also used to integrate a Ka-Band for active integrated antennas. Other integration techniques have also been proposed for Ka-Band MMIC packaging besides LTCC. In [6], an LCP package was used to package a two stage LNA in a single receiver front-end. Aihara in [7] also uses LCP to package a single Ka-Band front-end. He also packages a two stage LNA in a receiver front-end. Another approach has been to use a BeO ceramic with a Kovar housing [8]. This approach offers a very credible solution to the heat dissipation problem and retains excellent high-frequency properties. The technique, however, would lead to very high costs, especially where many carrier layers are necessary, and potential fabrication challenges like planarity of the carrier and sizes and tolerances of vertical interconnects and cavities.

In [9], the authors integrated GaAs die using MCPCB technology and showed that this technique could be suitable for K and Ka-band applications. This paper will further investigate a K-Band receiver front-end composed of four GaAs ICs that are built in MCPCB technology, where the dies are placed in cavities in high frequency laminates to provide adequate high frequency properties, and directly on the metal core to address thermal challenges of integrating a high density of dies.

The PCB material was selected based on processability, dielectric loss tangent and available thicknesses. The design of the PCB is shown in Fig. 2. The die have a thickness of 100μm and the die attach material will have an estimated thickness of 50μm. Therefore, a laminate and its top side metallisation should come as close to 150μm as possible. Ultimately Megtron6^™ was selected. With two 60mm prepreqs and a 35μm thickness metallisation on the top side, we achieve a very close thickness match of the dielectric stack at approximately 156μm. Megtron6^™ also has a low dielectric loss tangent, good processability, and wide availability.

The goal of the cavity structuring was to keep the wirebonds as small as possible while still allowing automatic die placement inside the cavities. Laser technology was selected, as opposed to milling technology, due to its improved tolerances of +/−50μm. Ideally the cavity would be structured as close to the outline of the front end as possible, to facilitate the shortest possible wirebonds. The chips have a length tolerance of +/−50μm. The placement tools require additional space of 50μm around the edge of the chips. Therefore, the cavities were structured with 100μm of additional space surrounding the expected sizes of the GaAs dies.

Fig. 3 shows the integration of a 20GHz receiver front-end, realized with the amplifier and frequency doubler HMC578, the TGP2615 phase shifter from Qorvo^™, the AMMC-6530 single sideband mixer from Avago^™, and the low noise amplifier (LNA) CHA3689 from United Monolithic Semiconductor^™.

In an active antenna, the front-end would be integrated into an array, where each front-end drives an antenna or group. To achieve the highest possible beam steering range, the antennas would require a spacing of λ/2. Fig. 3(a) shows the front-end length, relative to a half-wavelength and it can be seen that driving a single antenna would be unlikely because the front-end is only barely smaller than a half-wavelength.

To enable beam steering, each front-end would include a phase shifter. This phase shifter could be integrated either in the RF path (between the mixer and PA), in the IF path before the mixer, or in the LO path before the mixer. Due to availability of components, the front-end in this paper places the phase shifter in the LO path between the mixer and the LO driver amplifier. This disadvantage to this approach is the requirement of higher LO power to compensate for the losses in the phase shifter.

The mixer has a recommended drive power of 10dBm for optimal use. Below this level, it will not achieve its highest possible output power and it could potentially provide poorer isolation. The HMC576 is being driven with a 5dBm input power, which is within the 2-6dBm required by the component. A typical output power for the HMC576 is 17dBm. This is well above the 10dBm mixer drive power but the mixer, as well as the wirebonds, between the HMC576 and the mixer introduce losses. The phase shifter is expected to introduce 7dB losses and the wirebonds and additional 0.3dB. This leaves the drive power for the mixer slightly below what is recommended for the mixer. This means that the mixer output power will not be as high as expected and, even though it is a small difference, it could result in a difference of several dB between what is expected and what is observed at the IF output.

The front-end measurement setup is shown in Fig. 4. A 20GHz RF signal is generated with a PNA set to continuous wave output. This signal is connected with cables to a cascade GSG probe to a microstip line on the PCB, which is connected with wirebonds to the LNA. The LO generation is generated with a signal generator and is set to 8.6GHz. The IF signal leaves the mixer through transmission lines, GSG probes, and cables to an 90° hybrid coupler and is measured with a Agilent spectrum analyser. The output signal is expected to be 2.8GHz, which is the difference of the RF input frequency and twice the LO frequency.

Fig. 5. Shows the relationship between the IF output power dependent on the LO drive power and the RF input power. Starting at approximately −40dBm input RF power, we see a linear behavior at the IF output. The LO input power shows a maximum output power at approximately 5dBm.

Above it is calculated what had been expected during the measurement. The losses of the elements in the measurement setup were measured individually and the gain of the LNA was taken from the datasheet. Looking at Fig. 4(a), at a LO power level of 5dBm, there is an IF output power of 29.2dBm. This is a difference of 5.35dB from the expected power output. We expect that this difference comes from a lower LO drive power that is recommended by the mixer, which was discussed in the previous section.

A thermal camera was also used to examine the effectiveness of the metal core at spreading the heat generated by the chip. The results are shown in Fig. 6. The camera shows that the hot spots on the active chips (HMC578 and CHA3689) are approximately 7°C higher than the temperature otherwise on the chips. The temperature otherwise on the chips show little difference in temperature to the copper core directly near the active chips. The 20GHz receiver front-end is almost 10°C cooler than the 60GHz transmitter front-end because the LNA dissipates less power than the power amplifier.

This paper has investigated a K-Band receiver front-end composed of four GaAs ICs that are built in MCPCB technology. The dies are placed in cavities in high frequency laminates directly on a 1mm copper core. This approach is designed to provide adequate high frequency properties, and simultaneously to address thermal challenges of integrating a high density of dies. We have identified this integration approach as a viable concept for the integration of high power active antennas for high frequency satellite communications.