Power amplifiers: Getting to the G-band and beyond

BY ZACHARY GRIFFITH AND MIGUEL URTEAGA FROM TELEDYNE SCIENTIFIC

At the turn of the twenty-first century, engineers could grab the headlines with claims of a monolithic IC working at hundreds of gigahertz. Fast-forward fifteen years to today, however, and the fabrication of such a device doesn't even raise eyebrows "“ there are ICs operating within the G-band at 140 GHz and 220 GHz that are finding deployment in a wide range of applications.

What is special about both these frequencies "“ 140 GHz and 220 GHz "“ is that they are at low-loss operational "˜windows' in the Earth's atmosphere. Consequently, radiation at these frequencies can travel substantial distances, underpinning the construction of a wide variety of systems, including: very high data rate point-to-point links; personal imaging systems for detecting concealed weapons; synthetic aperture radar (SAR) systems operating with multiple frame rates per second, which can track manoeuvring targets and image scenes on the battlefield; and power amplifiers driving brethren based on vacuum electronics, and ultimately enabling sources of up to 100 W.

In the US, the Defense Advanced Research Projects Agency (DARPA) has supported the quest to develop G-band ICs by devising, funding and co-ordinating several programmes, including TFAST, SWIFT, FLARE, HiFIVE, and THz Electronics. Such efforts have provided the proof that InP, which has a higher electron transport velocity than silicon and SiGe, is well suited for the construction of G-band circuits with sufficient gain and bandwidth "“ and it is also capable of yielding terahertz transistors.

At Teledyne Scientific of Thousand Oaks, California, we have played our part in many of these DARPA programmes, developing during the last decade an InP-based double heterojunction bipolar transistor (DHBT) technology with high-frequency capability. Compared to its cousin, the InP HEMT, the DHBT has more than double the breakdown voltage, and it holds the key to fabricating state-of-the-art G-band power amplifiers. That is not the only merit of this class of transistor, however, as it also sports superior device uniformity, thanks to its vertical electron transport.

Laying the foundations

Our efforts at developing an InP-based G-band amplifier began with the development of the DHBT technology. This was not trivial, because it is not permissible to substitute InP into a SiGe HBT process. Instead, our engineers had to design and have grown an epitaxial structure with two separate heterojunctions: one between the InP collector and InGaAs base, and the other between the InGaAs base and InP emitter. In both cases, the junctions could not be abrupt, with alloy grading needed to eliminate conduction band discontinuities. Extensive studies led to the fabrication of such structures, which also had to provide to the semiconductor terminals contacts that are ohmic and exhibit a very low parasitic resistance.

To ensure adequate bandwidth and gain, delays associated with parasitics had to be minimised by adopting narrow transistor features. This was accomplished with recipes developed for e-beam lithography and i-line photolithography. Device fabrication followed, using 100 mm wafers and multi-level interconnects. Such efforts have enabled the construction of MMICs with a bandwidth exceeding 670 GHz, formed using the 1.1 THz f_max, 125 nm technology. This is a tremendous accomplishment that involved overcoming many challenges on the way to the fabrication of world class G-band and THz MMICs.

Development of these highly capable integrated circuits has included efforts to shrink transistor dimensions. This has spawned a portfolio of technologies: 500 nm transistors, our most mature technology capable of producing MMICs with thousands of DHBTs; 250 nm transistors, a mature technology for MMICs with upwards of 500 DHBTs; and, under development, transistors at a node with a 125 nm emitter that can be used to form MMICs with 50-100 DHBTs. Note that for G-band sources and frequencies up to 300 GHz, the 250 nm node has a great set of attributes, combining an f_t of more than 350 GHz with an f_max that tops 600 GHz and breakdown voltage of 4 V.

In order for this technology to be useful, there must be accurate predictive modelling of the transistor, passive components and the wiring environment. This has been realised by developing, through numerous iterations, a scalable large-signal model of the DHBT. Accurate predictions of the performance of this transistor over a range of currents and voltages are then possible; and in turn, this aids the designing of a power amplifier, because it is possible to calculate the optimal impedance to the DHBT for the highest output power.

Another requirement for high-performance MMICs operating in the G-band and beyond is to have low loss interconnects. To realise this, we use 1 Âµm-thick benzocyclobutene intermetal layer dielectric spacing with an electrical permittivity of 2.7 within the four-metal layer interconnect (see Figure 1 for an example of this technology). One downside of such a device is that its foundation, InP, is not as sturdy as silicon or GaAs. To address this we have a robust process for thinning 100 mm InP wafers to either 75 Âµm or 50 Âµm. We are able to incorporate thru-substrate vias and metallization, enabling backside grounding of the chip and suppression of parasitic modes at G-band and terahertz frequencies.

Figure 1. A scanning electron microscope cross-sectional image of the Teledyne 250 nm InP HBT and MMIC technology. MBE on 100 mm InP substrates forms epistructures featuring a highly doped, 30 nm-thick graded InGaAs base. High breakdown is ensured with a 150 nm InP collector that employs super-lattice grading between InGaAs and InP. High semiconductor doping at the collector, base, and emitter metal-semiconductor interfaces permits very low parasitic ohmic contact resistance "“ this is critical for achieving high gain and high f_max transistors.

Ensuring uniformity

The construction of precision analogue and comparator circuits, as well as highly combined power amplifiers, is only possible with MMICs that are formed from devices with identical DC and RF characteristics. Such a requirement plays into the hands of the HBT, which has two key advantages over the HEMT: the characteristics of the n-p-n transistor are well defined by epitaxial design, growth, and drift-diffusion electron transport theory; and electrons traverse vertically inside the device, and tend to be within the width of the emitter feature.

In contrast, electron transport in the HEMT is horizontal, with carriers traversing from source to drain via the channel. One implication is that for the construction of transistors that could be used in G-band and terahertz MMICs, gate feature sizes of only 20 nm have to be defined by lithography and metal deposition. Any variations in how the gate is formed, in the separation of gate electrode and channel, will modify HEMT performance and place an upper limit on the number of devices that can be used within such a high-frequency MMIC.

Thanks to the nature of the electron transport in the HBT, the accuracy of predictive and scalable large-signal models, which are based on the Agilent-ADS III-V HBT model, is superior to those for the InP HEMT (the latter are either empirical or small-signal models, which are compromised by limited bias range and accuracy). Having access to a highly predictable, large-signal model is incredibly valuable, because when a designer has to consider and potentially adopt a new technology to their MMIC, mixed-signal, and system needs, they can slash the time and cost of this project. These savings result from a trimming of design variations and a reduction in the number of fabrication runs.

Assessing performance

Transistors are often judged by their values for maximum oscillation frequency f_max and the breakdown voltage, but for MMIC designers, other figures are more useful. More valuable is the gain that the DHBT has in common-emitter and common-base configuration at a given operating frequency, and the on-state destructive voltage at different operating currents. These values are much better than those for SiGe HBTs and InP HEMTs (see Figure 2). Common-emitter characteristics show a peak current density of 2.5 A/mm (10 mA/µm²) at a 0.8 V knee voltage, while at low-voltage operation, the device may be operated up to a power density of 4 W/mm. At a 4 V bias, a DC current of 0.75 A/mm is feasible.

Figure 2.  Measured DC and modelled RF performance of the Teledyne 250 nm InP HBT technology for 220 GHz PA applications.

A noteworthy feature in these plots is how the gain varies with the configuration employed for the transistor. In common-emitter configuration at 220 GHz, plots of gain have a peak of 7.5 dB, and much lower values at the power amplifier load-line endpoints. Turn to a common-base configuration, however, and gain is 9dB at the operating endpoints, and peaks at 12.5 dB. This performance, which is superior to that of the SiGe HBT and InP HEMT, is partly due to very high collector-to-emitter RF isolation. In addition, there is an absence of Miller multiplication of the base-collector junction capacitance, which does play a role when the HBT is common-emitter configured.

Before MMICs operating at these G-band and terahertz frequencies can be designed, engineers need to have in their toolkit a low-loss interconnect environment and passive components. One issue with the 250 nm InP HBT, which has very high gain at G-band, is that it is prone to instability when the traditional design approaches and interconnect schemes used for Ka-band to W-band MMICs are employed. Our solution is to turn to thin-film microstrip wiring. Compared to grounded coplanar waveguide and thru-substrate microstrip, this approach has four advantages: there is lower loss; there are more interconnect layers, which can be spaced much more densely; there is no need for wafer thinning and dicing prior to MMIC evaluation; and the accuracy of electromagnetic modelling is higher, thanks to tremendous simplification of the modelling of the ground-return currents.

Our thin-film microstrip technology employs either the top-most or bottom-most metal layer as the ground-plane, with the remaining metal layers used for DC and RF interconnect routing. With this approach, measured insertion loss is just 1.1dB/mm at 220 GHz for a 50 Ω transmission line.

Armed with this knowledge of the capability of this technology, combined with very accurate modelling of the transistor, we have built a range of 220 GHz MMICs and PAs. One of their features is that they avoid the traditional approach for making signal sources that are needed to construct transmitters and receivers. The conventional, effective approach for generating an LO source at these frequencies is to multiply up the frequency reference. Unfortunately, however, multiple chips are needed to incorporate this LO into a transmitter or receiver architecture "“ and using multiple chips adds to cost and introduces complexity. There are several reasons for this, including the need to evaluate each chip, and the need to machine multiple waveguide blocks or channels and load chips. In addition, the waveguide block has to be re-evaluated prior to shipping the final transmitter or receiver assembly.

We avoid these issues with our 250 nm InP HBT technology. With this, we make a monolithic single chip receiver or transmitter that contains a PLL frequency source, mixer, and either LNA (receiver) or PA (transmitter). The most sophisticated part of this is the PLL circuit, which employs a phase-detector, active loop filter, voltage-controlled oscillator (VCO), and 2:1 dynamic divider (see Figure 3 for details).

Figure 3.  Teledyne's 220 GHz phase-locked loop is formed with 250 nm InP HBT technology and has a 220.0 to 225.9 GHz locking range. The output from the 220 GHz voltage-controlled oscillator drives a 2:1 dynamic divider to supply a fifth-order sub-harmonic phase detector a 110 GHz signal, using a 22 GHz reference clock.

An advantage of this PLL approach is that it does not require an additional frequency division of the (VCO) signal to the phase detector. Thanks to this, power dissipation falls, and there is also a reduction in MMIC transistor count. Operation of this part of the circuit involves feeding the phase-detector output into the active loop filter, with the output of this supplied to the VCO control node to phase-lock the VCO operation. Note that one implication of the inverted thin-film microstrip is that the ground plane conceals the circuit blocks "“ labels show where they reside.

Measurements of the output of the PLL reveal a centre frequency of 223 GHz, which is close to the simulated value of 220 GHz. Phase noise at 220 GHz, 223 GHz, and 225.9 GHz is very similar, with values for 10 kHz and 100 kHz offset of -61 dBc and -83 dBc. The PLL may be integrated, on-chip, with a mixer and an amplifier to construct a transmitter or receiver. Such a chip requires just one waveguide block, making it much cheaper than products based on multi-chip, frequency-multiplication approaches.

High-frequency amplification

The development of our high-frequency InP DHBT technology has been applied to high-power PAs operating at 220 GHz and the upper end of the WR04 frequency band (it extends from 170 GHz to 260 GHz). These amplifiers produce far more power than those formed from either InP HEMTs or large diode multiplication units, an advantage that is highly valued at the system level. Thanks to packaging advances at G-band frequencies, modules can be constructed from a battalion of PAs, increasing the RF power. The key to realising this is being able to produce and select MMICs with identical gain, phase, and saturated output power.

By avoiding the need for wafer thinning prior to RF evaluation, we can carry out automated wafer probing of the PAs. Thanks to advances in high-frequency wafer probes from GGB Inc., more than 11,000 PAs can be evaluated before the probe-tips require maintenance. The probing determines values for S-parameters, and with software sifting through this data, just a "˜spot-check' of large-signal performance is needed. This is a tremendous breakthrough, because testing can account for up to 40 percent of the total cost of production for MMICs operating in the G-band and terahertz range. With our approach, the costs associated with testing are far less than this.

We have built our 220 GHz PA using a cascode topology. We arrived at this decision after considering the footprint of the HBT PA cell and the wiring layout. By taking this approach, we can ensure that the PA is stable and will allow an increase in power by combining monolithically up to 16 PA cells on a single chip. This level of integration is possible because interconnect loss is only 1.1 dB/mm, while the 2:1 and 4:1 combiners have insertion losses of just 0.45 dB and 0.5 dB, respectively (see Figure 4 for an example, a three-stage amplifier based on four cells).

Figure 4. The cascode power cell (top) and a block diagram of the three-stage, 4-cell output combined PA (bottom). This three-stage, 4-cell combined amplifier is the basis for Teledyne's 8- and 16-cell combined amplifiers, where the 2:1 combiner structure is used to add them together.

Our three-stage, 4-cell amplifier delivers a very impressive level of performance, including a gain in excess of 28 dB from 205-271 GHz (see Figure 5). At 220 GHz, this gain is a record-breaking 35 dB. Output power peaks at 80 mW at 220 GHz; and it is 56.0 mW at 250 GHz and 46.5 mW at 260 GHz "“ these are the first reported data points greater than 15 mW for a solid-state PA at these frequencies.

Figure 5.  The RF performance of Teledyne's high-bandwidth, three-stage, 190-260 GHz PA. Gain, determined by the S-parameter S₂₁, is greater than 28 dB from 205-271 GHz. Large-signal data shows output power and gain across frequency, where the PA input power is constant at only -3.0 dBm (0.5 mW P_in). Output power is greater than 50 mW from 190.8-254 GHz, and hits 80 mW at 220 GHz.

Even higher powers are possible with a three-stage amplifier formed from 16 cells, made from a stack of four 4-PA cells united with 2:1 combiners. Creating a MMIC using 1.41 mm of HBT led to the highest power G-band 220 GHz MMIC demonstrated to date. Output is 208.7 mW at 210 GHz, and while a power-added-efficiency of 3.4 percent may seem low, it is respectable given that at this high output power the large-signal gain of the amplifier exceeds 13 dB (see Figure 6 for details of performance, and Figure 7 for a summary of the output powers of 4-, 8-, and 16-PA cell combined amplifiers targeting 220 GHz).

Figure 6. Teledyne's high-gain InP HBT PA can deliver more than 200 mW. Formed by taking three-stage 4-PA cell MMICs and stacking them for a total of 16-PA cells "“ and combining the outputs using 2:1 combiners "“ this is the highest power G-band 220 GHz MMIC ever. The S₂₁ mid-band gain is 26.3 dB from 205-243 GHz, and at 200 GHz, output peaks at 220 mW.

Figure 7.  A performance summary of output power across frequency for Teledyne's 250 nm InP HBT WR04-band solid-state PAs.

Thanks to DARPA's sponsorship, we are now seeing our 220 GHz amplifiers based on 250 nm InP DHBT technology leaving the lab in volume levels to construct Watt-level solid-state power amplifier modules operating at 210-230 GHz (see Figure 8). This milestone, and the "˜productization' of the G-band MMIC, indicates that the InP HBT is well-positioned to address MMIC requirements that will spur commonplace G-band and terahertz systems.

Figure 8.  A performance summary of state-of-the-art PAs at the MMIC die and module-level. The modules from Raytheon and Nuvotronics will be presented as part of the technical program for the International Microwave Symposium 2015, held in Phoenix, AZ.

-The views expressed here are those of the authors and do not reflect the official policy or position of the Defense Advanced Research Projects Agency or the United States Department of Defense. The work presented here was sponsored by DARPA under the following programs:  TFAST (N66001-02-C-8080), FLARE (N66001-02-C-2025), HiFIVE (W911NF-08-C-0500), THz Electronics (HR0011-09-C-060), Mobile HotSpots (FA8750-12-C-0279), as well as HR0011-13-C-0013 and HR0011-13-C-0015. The authors sincerely thank all of the Program Managers of these programs for their support.