NGST transforms InP transistor manufacture

The company's InP HBTs have been used to build efficient, linear power amplifiers targeting the cellphone market, and its HEMTs have been incorporated into W-band low-noise amplifiers (LNAs) for improving the performance of passive millimeter-wave (PMMW) imagers that will be used to detect hidden weapons, for instance, at airport security.

NGST's Richard Lai says that InP devices offer several advantages over those made from materials such as GaAs, SiGe and silicon. These include superior gain, low noise (see figure 1), low power consumption and high efficiency, as well as the superior thermal conductivity that arises from the InP substrate.

Although InP transistor production is less mature than that of GaAs, the microelectronics product manager says that for RF devices, NGST's InP yield exceeds that of its GaAs transistors. This results from wider margins on RF performance set for InP devices. Because NGST makes the GaAs and InP transistors on the same 100 mm line, the devices have similar fabrication costs, and the only real drawback of InP is the relatively high substrate price.

InP HEMTs and HBTs are grown by solid-source MBE on a multi-wafer platform accommodating up to seven 100 mm substrates. The subsequent device fabrication shares many of the passive component and backside processes used for GaAs HEMT and HBT production, improving reproducibility and cutting costs.

NGST's HEMT and HBT processes use the same steps for nitride deposition, thin-film resistor formation, interconnect metal deposition, and backside processing. The only major differences are in the etch chemistries used (and related resist processes), the metals deposited, and the anneal conditions.

Lai believes that a good via etch process holds the key to the manufacturing of both InP HEMTs and HBTs. NGST's approach is to bond 100 mm InP substrates to 109 mm sapphire wafers, before mechanically grinding the InP substrate down to a thickness of either 50 or 75 μm. The through-substrate vias are then formed by reactive ion etching with a Unaxis system that can handle three wafers simultaneously. This process, which takes 90 min, produces etch uniformities below 3% across a three-wafer batch and a 100% via yield.

According to Lai, NGST's InP HBTs operate at double the frequency of their GaAs counterparts when using only 50% of the bias voltage. The InP device's higher-frequency performance results from its smaller, 0.8 μm-wide emitter. The company is also currently developing transistors with emitter widths of less than 0.15 μm that can operate at over 300 GHz.

NGST's InP HBT production line has an 85-90% device yield. Standard characterization tools such as bulk resistivity and X-ray diffraction measurements determine epiwafer quality, and ensure a repeatable growth process. In addition, thinner structures are grown twice a week to confirm layer thickness, doping and junction quality.

NGST produces two different HBT designs. One variant, targeting digital applications, uses a thin base and collector to deliver high-speed performance at a moderate breakdown voltage, and has values of 150 GHz for both ft and fmax. The second HBT design is suitable for microwave power applications, and its thicker base and collector layers deliver a larger breakdown voltage and improved linearity. However, this device's high-frequency characteristics are inferior, with ft and fmax values of only 80 GHz and 130 GHz respectively.

Lai believes that NGST's InP HEMTs could be used to produce cheap, highly linear and highly efficient amplifiers for the cellphone market. The company has built a "power cell" to operate at 2.4 GHz that uses 1.5 × 30 μm transistors. This module has a power- added-efficiency (PAE) of 63.4%, and 20.8 dB gain at a current density of 31.9 kA/cm2. The technology has also been deployed in a single-stage MMIC with output power of 500  mW and a PAE of 50% at 1.95 GHz.

HEMT production on 100 mm substrates, which also boasts an 85-90% line yield, is an extension of NGST's space-qualified 75 mm process. Growth time for these structures is typically just 30 min, which is 4-6 times faster than that for GaAs HEMTs requiring the insertion of a metamorphic buffer layer. The higher wafer throughput and the lower labor-cost per wafer are claimed to offset InP's higher substrate costs.

Lai predicts that InP HEMT manufacturing costs will fall below those associated with their GaAs counterparts, so long as both transistor types are fabricated on the same-sized substrates, and that the cost of InP substrates falls as projected. He also points out that the company's MBE equipment and process line can accommodate 150 mm InP wafers with minimal tooling adjustments, although sufficient quantities of high-quality 150 mm InP substrates have yet to become available.

NGST produces InP HEMTs with an In0.6Ga0.4As channel and relatively good noise characteristics, and higher-power, higher-efficiency versions with an In0.53Ga0.47As channel. Both can be produced with 0.15, 0.l or 0.07 μm gate lengths, depending on the required frequency performance.

These InP HEMTs can be used in LNAs for millimeter-wave imagers operating in the W-band (75-110 GHz). According to Lai, InP HEMTs can increase camera sensitivity by three to five times, decrease power consumption by the same amount, and deliver a two- to three-fold size reduction. The LNAs produced by NGST have already been installed in millimeter-wave receivers in telescopes operating throughout the world.

Lai adds that because of the threat of terrorism, there is now a demand for millimeter-wave imagers with larger focal arrays that can provide sufficient detail for easy detection of concealed weapons. This will lead to increased efforts to reduce the imager's cost, size, weight and power consumption. He believes that these detectors will benefit from the high performance and reasonable price of LNAs based on InP HEMTs. Detector size and weight could also be reduced by increasing the HEMT operating frequency to 140 or 220 GHz, says Lai.

For many years InP devices have promised low noise with excellent performance at higher frequencies, but they have always been associated with higher costs. NGST's new 100 mm process suggests that InP device manufacture can approach thatof GaAs, and that InP transistors will soon become affordable components used in next-generation cell- phones and millimeter-wave cameras.