The US Department of Defense views GaN MMICs as too expensive and insufficiently reliable for its needs. To address these shortcomings, it is spearheading a project to drive down cycle times, increase yield and improve reliability. Richard Stevenson reports.

GaN has great characteristics for making RF transistors. This material can be used to build devices that are head-and-shoulders above those which are made from GaAs: They deliver far higher efficiencies, produce far higher power densities and bandwidths, and are able to operate at much higher temperatures.

This wonderful set of attributes has not escaped the notice of the US military, which views this wide bandgap transistor as a very promising component for many aspects of electronic warfare, including electronic attack, electronic protection and electronic warfare support. For example, GaN RF chips could be used in the construction of phased array antennas, which will be deployed in radar and electronic systems.

To aid the development of GaN RF devices for military applications, the Defense Advanced Research Projects Agency (DARPA) has funded the development of this wide bandgap technology through initiatives such as the Wide Bandgap Semiconductor for RF Applications program. Efforts such as this are focused on technology development, and they are typically high-risk, high pay-off investments. In the case of GaN, they have borne much fruit, so the government services – the Navy, Air Force and Army – are now keen to also help to take the maturity of GaN RF technology to a new level.

To do this, the Department of Defense (DoD) is funding a Title III program for GaN RF technology. The official goal for the latest effort – which follows Title III programs on silicon-on-sapphire, GaAs wafers, InP, and SiC wafers and MMICs – is to instantiate a production capability defined as a manufacturing readiness level of 8: This means that the process is ready for low rate initial production in a DoD acquisition programme.

“[The project] is considered a success if the production processes are capable of producing devices of sufficient yield, cost, performance and reliability to be advantageous to both defence and commercial customers,” explains the programme’s manager, Gene Himes. He knows that the production capacity/capability enabled by the Title III program should be self-sustaining long after the project is over. To try and put this into place, the project will address the high cost of GaN devices and their reliability, and involve measures to increase the yield, throughput and capacity of GaN lines in US companies.

The big four

Three of the biggest US chipmakers that already have GaN production capability are currently involved in the Title III program, while a fourth is just starting out. Raytheon is running an X-band (7-11.2 GHz) project that started in 2010; Cree and TriQuint are engaged in independent programmes for S-band (2-4 GHz) and wideband (6-18 GHz) MMICs that kicked off in 2011 and 2010, respectively; and there is also a Ku-band (7-11 GHz) MMIC project by Northrop Grumman that starts this year.

Insights into the progress of the first three of these programmes were offered at the CS Mantech conference held in New Orleans from 13 May to 16 May 2013.

At this meeting, Cree Program Manager Ryan Fury detailed improvements to the company’s 75 W S-band and 25 W wideband die, which are fabricated with the company’s G28V3 and G28V4 HEMT processes, respectively. Cree introduced the latter process in 2007 to target a variety of DoD and commercial applications.

“The first volume applications were for jammers,” says Fury, who explains that Cree’s GaN devices were used in products sent to Iraq and Afghanistan to prevent radio-controlled detonation of improvised explosive devices (IEDs). “Over 300,000 high performance, multi-stage GaN MMIC power amplifiers were sold using our G28V3 process,” reveals Fury, who adds that more recently the process has been used to make products for military radar, cable TV and telecom infrastructure.

Cree has recently introduced a higher voltage variant of the G28V3 process, the G50V3 – it is designed for 50 V operation. “This process provides almost twice the power density of our baseline 28 Volt process, and is ideally suited for high power radar and low cost telecom infrastructure applications, where either high power or low cost – in terms of dollars-per-Watt of RF output power – are critically important.”

In 2012, the Durham-headquartered chipmaker expanded its GaN portfolio even further, adding a G28V4 process to plug the gap between 6 GHz and 18 GHz that was not addressed by the 0.4 µm G28V3 processes. This higher-frequency process can be used to make products for X-band radar, electronic warfare jammers and satellite communication systems.

“We’ve also introduced a higher-voltage version of the X/Ku-band process, the G40V4, which operates at 40 volts,” says Fury. “ This is ideally suited for high frequency, pulsed-power applications, such as travelling-wave tube replacements.”

The major difference between the V3 and V4 processes is the gate length: It is 0.4 µm in the former process, but in the latter it is just 0.25 µm. Standard optical lithography and a single SiN etch is used to realise the 0.4 µm gate length, while the shorter variant is fabricated with optical lithography and a standard sidewall spacer. Source-connected field plates are employed to increase gain and RF power density. Using a second field-plate also trims peak fields within the transistor, so this device can last longer. Thanks to the use of optimised field plates, the 1 mA/mm breakdown voltage of these HEMTs exceeds 120 V.

At the beginning of the Title III program, Cree’s engineers carried out a baseline assessment of the production cost, yield, cycle time and performance of their S-band and EW-band MMICs, which were formed by depositing an insulating GaN buffer and AlN and AlGaN cap layers on 100 mm semi-insulating SiC substrates by MOCVD. This initial study showed that Cree’s process exceeded the baseline key performance parameters set out by the title III program, with yields up to 75 percent above the benchmark, and cycle times 42 percent quicker than the initial standard.

During the programme, efforts have focused on driving yield higher, trimming cycle times and reducing manufacturing costs. This has included work to optimise gate metal electrode lift-off and reduce the number of damage sites; a switch to a new post-backside de-bond process that cuts front-side damage and drives down cycle time; and the qualification of a SiN passivation tool from a manually loaded, older PECVD platform to an automated version that diminishes particle count and reduces handling damage.

These efforts have increased yield, while cutting cycle time and cost. S-band and EW-band yields are 8 percent and 18 percent above the mid-point goals for the programme, while the MMIC cost and cycle time are 36 percent and 25 percent below the interim benchmark (see Figure 1).


Figure 1: Through process improvements, such as refinements to the gate metal lift-off procedure and the post-backside de-bond process, Cree’s engineers have reduced the cycle time of their manufacturing process and increased its yield

Cost savings have been realised, thanks to cutting material costs, increasing automation and boosting throughput. “Higher factory loading is primarily due to the rapid adoption of our GaN products for high-volume markets, such as telecom base stations,” explains Fury. “These cost savings are helping us to further increase GaN penetration into the military and commercial markets.”

High reliability is another target within the Title III program. For continuous operation at a temperature of 125 °C for the back of the MMIC die – which equates to a maximum junction temperature of 225 °C – the mean-time-to-failure (MTTF) goal is a million hours. Cree’s MMICs are far more reliable than that, with MTTF values of 55 million hours for the G28V3 process and 150 million hours for the G28V4 process. “Our GaN process has the highest rated operational channel temperature of any currently on the market,” claims Fury.

It might seem that with goals appearing to be met so easily, these targets should have been higher. But that misses the point. “The goals were structured to ensure that Cree would be able to produce GaN products with the cost, performance and reliability needed to support a number of critical US defence programmes,” says Fury.

One key challenge remains for Cree in this Title III program: Passing an 8000-hour, RF high-temperature operating life test. This is designed to provide an additional assessment of the robustness of the process.

Cree is now involved in the final assessment phase of the programme. However, its processes have been independently assessed by two different external customers. They have judged that it is has reached a manufacturing readiness level of 8. In addition, the company is using the lessons it has learnt form the programme as it is developing of a high frequency,

Ka-band (26.5 GHz to 40 GHz) process, known as V5. “This will employ a 0.15 µm gate length using the same robust optical lithography process that has been improved on this programme,” says Fury.



The Title III program will help to reduce the cost and increase the reliability of GaN MMICs that can be used for X-band radar systems. Examples of this type of system include the sea-based X-band radar, which can provide highly advanced ballistic missile detection to discriminate a hostile warhead from decoys and countermeasures. Credit: US Navy 


TriQuint’s approach

Like Cree, TriQuint Semiconductor has long history in the development and manufacturing of GaN MMICs. Its efforts in this area date back to 1999, and nearly a decade later, in 2008, it released a qualified manufacturing process for GaN MMICs on 3-inch SiC substrates. Two years after this, the company released a 4-inch process, which is used for standard product development and foundry projects for external customers.

Efforts in the Title III program have revolved around two devices that have been designed for high power and power-added efficiency: an S-band MMIC, which is a two-stage power amplifier; and a wideband PA that covers the X-band and the Ku-bands. Both amplifiers are fabricated from HEMTs formed on 100 mm, c-plane 6H and 4H semi-insulating SiC wafers featuring nitride epilayers deposited by MOCVD. Electron-beam patterning defines the 0.25 µm gate length, with devices sporting a source-connected field plate to reduce high-field-related device degradation.

During the improve-and-refine phase of the programme, TriQuint’s engineers have focused on tightening manufacturing process variability; implementing GaN-specific characterisation and tracking, as well as cutting cycle time.

“The refinements we have undertaken are based on well-developed, design-of-experiment techniques, since multiple material and process interactions can occur,” explains Douglas Reep, Senior Director of Research in the Infrastructure and Defense Products division at TriQuint. “Resolving issues isn’t always easy, but when statistical techniques are employed, good manufacturing solutions are what we expect.”



Cree has supplied the US army with over 300,000 multi-stage GaN MMICs, which have been used to build systems that can jam signals for setting-off improvised explosive devices. Credit: US Army

This approach uncovered a source of a wafer-to-wafer dependence associated with the nitride capacitor deposition process. This variation stemmed from a coupling of the wafer to the plasma deposition system, and it was eliminated by adjusting one of the process steps.

Another area where TriQuint has improved is tightening variations in gate leakage, which can be impacted by the condition of the surface. Engineers looked at the impact of changes to process and tool, and have moved to equipment with automation, which improves process monitoring, and implemented a new process that reduces defects.

Resist removal procedures were also scrutinized at TriQuint. Improvements in physical cleanliness led to a substantial fall in defects, but these refinements at individual steps in the process failed to tighten variations in electrical characteristics. However, when the new processes were used throughout the GaN flow, they drove a significant reduction in variability.

One possible cause for variation in devices is differences in substrate quality. Traditional optical inspection is not particularly useful, because it cannot detect scratches, due to the transparency of SiC. To address this, and also provide a means of categorising defects such as micropipes, scratches and particles, TriQuint has invested in a Candela inspection tool. The engineers are in the process of establishing correlation between device failures and incoming material defects.

Cycle times have been reduced during the Title III program. Throughput of a SiC/GaN via etch step has been aided by the qualification of a second etch tool, and cycle times have been slashed by a factor of three by increasing the via etch rate. This involved optimising power and pressure, and also led to improved uniformity. Switching from e-beam to optical lithography could lead to further reductions in cycle time. “We have several processes that employ only optical lithography, and we use e-beam where we are pushing for performance growth and flexibility. For example, several of the new products we announced at IMS are manufactured using our latest TQGaN15 technology, which uses e-beam lithography to define its 0.15 µm gates,” explains Reep.

Reliability has also been assessed in TriQuint’s Title III program. The MTTF is greater than 10 million hours at a channel temperature of 200 °C, and in excess of 1 million hours at 225 °C. Like Cree, results on the 8,000-hour operational life are still to be reported.

There are 10 manufacturing readiness levels associated with the Title III program

Raytheon’s refinements

Raytheon’s engineers started work on their Title III program back in 2009, and have now realised the primary goal, demonstrating and validating manufacturing readiness level 8 and qualifying this process for insertion into DoD production systems.

This project involved three major tasks: processing 24 wafers with the baseline production process to evaluate yield and reliability; reducing variations through identification of root causes, using a minimum of 150 wafers; and processing 24 wafers with the improved process, to demonstrate that the requirements for manufacturing readiness level 8 have been satisfied.

For the first part of this project, 24 wafers were processed with a mask set that included FETs, small MMICs for reliability evaluation and large periphery X-band MMICs for yield evaluation. Reasonable yields were obtained for the latter device, but they were less than those for comparably sized GaAs MMICs. Meanwhile, reliability assessments based on three-temperature, DC Arhenius testing fulfilled the programme objective for MTTF, but a small proportion of the devices under test exhibited early degradation at 380 °C. This initial study shaped the primary goals for the next phase of the project: To improve X-band MMIC yield to a level comparable to that for GaAs MMICs, and to eliminate early degradation observed in a small proportion of devices at elevated temperatures.

Success resulted from increasing the yield of capacitors and improving the gate process, which largely addressed the early degradation issue. Engineers also investigated more aggressive DC and RF screening tests. More stringent screening reduced yield at the wafer level, but was beneficial elsewhere, leading to dramatic improvements in the predictability of reliability data and next-level assembly yield.

An assessment of the RF operating life of the devices was made at this point in the project, with the performance of seven X-band MMICs scrutinised during a 5,000-hour test, where they were driven in continuous-wave operation. No discernable change in power could be seen during this evaluation.

By the end of the project, GaN process yield more than tripled to a level that is comparable with Raytheon’s GaAs process. This gain contributed to a fall of greater than 75 percent in the cost-per-Watt associated with GaN, making it three times cheaper than GaAs, according to calculations that include substrate costs.

Reliability assessments at the end of the project show a MTTF at 150 °C of greater than a billion hours, more than three orders of magnitude higher than the goal set out in the programme. Even the time to 1 percent failure is more than 10 million hours.

Himes is pleased with the progress made by Raytheon, and also that accomplished by Cree and TriQuint: “Each of our GaN projects has demonstrated tremendous success in establishing mature and capable production processes.” He attributes this success to outstanding industry-government partnerships involving talented, committed industry teams that communicate bi-weekly with their government partners.