MMIC semiconductor tradeoffs

Updated May 27, 2007

Don't get too used to this page, it is going to get split up soon enough!

Click here to go to our page on MMIC suppliers

Let's start with a Microwaves101 rule of thumb!

Any microwave semiconductor house that doesn't invest in new technology, is going to go out of business in the long run. By long run, we mean five years.

At just about every meeting involving an empty suit, someone will ask "what's the difference between a SiGe HBT and a GaAs PHEMT MMIC?" Plenty! You can look like an expert and make your friends and enemies jealous of your knowledge if you study this page carefully.

Attention corporate spies, the information compiled here is widely known throughout the universe, so don't think we are giving away any proprietary foundry information here, from any sources. We just know what we scrape off the web…

Attention MMIC and discrete semiconductor foundries... any one of these topics is fair game for you to sponsor on your own private Microwaves101 page! And we'll become your new best friend, we promise! Attention MMIC nerds! Contact us if you want to write about one of these topics, for cold hard cash, so we can speed up the process of making this into a more useful resource!

Here are the semiconductor technologies that we review on this page:

GaAs MESFET

GaAs PHEMT

GaAs MHEMT

GaAs HBT

GaAs VPIN diode

Indium phosphide HEMT

Indium phosphide HBT

Silicon CMOS

Silicon LDMOS (separate page)

Silicon carbide LDMOS

Silicon germanium HBT

GaN HEMT

Antimonide based compound semiconductors (ABCS)

Horizontal or vertical?

Before we get too far into this subject, let's discuss the difference between vertical and horizontal semiconductors. HBTs and PIN diodes are vertical structures, each of the regions are grown in layers using some type of epitaxy. FETs of all kinds are horizontal structures. When we talk about vertical structures, where the "magic" takes place is referred to as the junction. For horizontal transistors (FETs) it's called the channel. So don't talk about the "junction temperature" of a PHEMT, or you will sound like an idiot!

What's a compound semiconductor?

A compound semiconductor is one where the crystal lattice uses two or more types of atoms. This is the case of gallium arsenide, gallium nitride and silicon germanium.

GaAs MESFET

Gallium arsenide MESFET was the original answer to "how can we make amplifiers at microwave frequencies?" The first GaAs MMICs demonstrated in the 1970s. Including HEMT and HBT technologies, literally billions of dollars have been spent extending fmax of GaAs products up into 100s of GHz.

The semi-insulating properties of GaAs substrates and the 12.9 dielectric constant make it an EXCELLENT media for microstrip or CPW design. It operates reliably up to 150C channel temperature. It is "radiation hard" for space applications. GaAs substrates are available up to six inches (150 mm) in diameter, which has been a long development since the first 2-inch wafers were available in the late 1970s. Sadly, GaAs MESFET MMICs will NEVER be cheaper than silicon, due to the starting material cost ($100s of dollars). GaAs parts are more fragile than silicon, and the thermal dissipation factor is not that good. GaAs MESFETs may be extinct in five years, because it doesn't cost much more to fabricate PHEMT or MHEMT on GaAs, and these technologies offer higher performance.

Examples

M/A-COM Roanoke foundry, TriQuint Oregon

GaAs PHEMT

GaAs PHEMT was the second MMIC technology to be perfected, in the 1990s. Breakdown voltages of PHEMT up to 16 volts make high-power/high efficiency amps possible, and noise figure of tenths of a dB at X-band means great LNAs, and made the DISH network possible, you lucky dogs!

PHEMT stands for pseudomorphic high electron mobility transistor. "Pseudomorphic" implies that the semiconductor is not just GaAs, perhaps AlGaAs/InGaAs/GaAs or some other secret recipe of 11 herbs and spices. Here's some further info on the the use of pseudomorphic in this context (sent in by some M101 fans!)

Actually, "pseudomorphic" means that the hetero layers are thin enough not to keep their own crystal lattice structure, but assume the structure (lattice constants especially) of surrounding material (lots of stress is involved),

If you look at a two dimensional cross section of the layer, you'll see that while it assumes the lattice constant of the bulk structure in the X direction, it tries to keep its original lattice constant in the vertical direction. This layer is indeed strained. For a GaAs pHEMT, indium is added to improve mobility and form a quantum well. Indium wants to growth the lattice and the typical range for useful thicknesses would be 10-25% on GaAs. You can also do strain compensation with the Schottky or cap layer.

The purist nerds of semiconductors often capitalize "PHEMT" as pHEMT. To them we offer this advice: get over it, or we will beat you up like we used to do on the playground, remember?

Examples:

TriQuint Texas, Velocium,

GaAs MHEMT

Recent work on metamorphic MHEMT has made premium InP HEMT performance possible (amps up at 100 GHz) at the same price as "regular" GaAs PHEMT. You can get noise figure and fmax equal to indium phosphide by using MHEMT, if you use a reputable foundry and indium content is high. You can actually exceed InP RF performance with indium content greater than 55%! The down side to all that indium is reduced operating voltage.

MHEMT stands for metamorphic high-electron mobility transistor. The channel material is InGaAs. "Metamorphic" implies that the lattice structure of GaAs is buffered using epitaxial layers to gradually transform the lattice constant so it lines up with InGaAs. InGaAs is normally grown on InP, which is expensive and fragile compared to GaAs. "Metamorphic" is changing the lattice constant by bond breaking as opposed to "pseudomorphic" which means just straining the heck out of it!

Examples:

BAE, Win Semi

GaAs HBT

The heterojunction bipolar transistor (HBT) is a new development, and can decrease the cost of GaAs amplifier products because the emitters are formed optically. GaAs HBT devices operate vertically, compared to the horizontal operation of FETs. However, for very high frequency, the emitter size must be made quite small, and the InGaAs layer is thick and is a thermal insulator, so these devices tend to run HOT. Typical HBT amps are "gain blocks", used in the UHF to C-band frequency ranges.

Typical supporters of HBTs will tell you that wafer yield up to 99% is possible.

Here's a great HBT paper written a few years ago by a Marconi employee. He's probably studying to get his real estate license now, while collecting unemployment compensation!

Examples

WJ

GaAs VPIN diode

PIN diodes make great switching elements. Vertical PINs (VPINs) are offered on some MMICs, but this is truly a niche market. As far as we know, nobody offers VPIN diodes and amplifier devices such as FETs on the same wafer.

Examples:

M/A-COM, TriQuint Texas

Indium phosphide (InP) HEMT

Indium phosphide HEMT has broken all of the upper frequency records, on the way to terahertz devices. However, there are serious drawbacks to this technology, not the least of which is its high cost. For this reason, InP is more regarded as a lab curiosity rather than a production process.

The actual semiconductor that is doing the work in so-called InP is actually InGaAs. Indium phosphide is merely the substrate that it is grown onto. The reason for this is that InGaAs shares the same lattice constant with InP, 5.87 angstroms.

InP substrates are small (3" typical, 4" are available but remember bigger is not always better when something is brittle). ER=12.4, close to that of GaAs. A huge drawback of indium phosphide technology is that InP wafers are extremely brittle compared to other semiconductors. Try shipping an InP wafer sometime. Silicon is the least brittle, and GaAs is somewhere in the middle.

Examples:

Velocium/Northrop Grumman

Indium phosphide (InP) HBT

Some people think that InP will have a second chance to become the most ubiquitous power amplifier technology for cell phones when new higher power density/lower voltage lithium ion batteries become available, as suggested in this December 2006 High Frequency Electronics article by Michael Gaynor. InP has superior low voltage performance compared to GaAs HBT.

Silicon CMOS

Silicon is so cheap you can make your roof out of it. It comes in 12 inch (300 mm) and bigger wafers. Processing is cheap. But it is not a good media for microstrip (lossy), and it is not rad hard. Silicon by itself doesn't make very good amplifiers above maybe X-band. Noise figure, power, are all second class to any of the compound semiconductors. It can only operate reliably up to 110C, but silicon is an pretty good heat dissipater.

Coming soon: a discussion of silicon-on-insulator (SOI!)

Here's a page on silicon LDMOS!

Silicon carbide LDMOS

Laterally-diffused metal oxide semiconductor technology, used to make power amps. The Freescale web site claims they pioneered the technology. Can withstand 200C channel temperatures. Good to 3 GHz, 10 watts.

Silicon germanium HBT

SiGe is a new development (in the last five years), and was originally predicted to put all forms of GaAs into the history books. SiGe can make very cheap parts, with performance maybe into millimeterwave, and processing on eight-inch (200 mm) diameters wafers. But the devices are not as high-performance as GaAs, in terms of noise figure and power. The setup charge at IBM to make a mask set is enormous, because 200 mm contact masks are needed (GaAs usually uses a 10X wafer stepper, these glass reticles are relatively cheap). You might pay $250,000 for that first SiGe wafer, but your one-millionth amplifier will be oh-so-cheap!

The poor insulating properties of a silicon substrate means it's not a good media for microstrip, so you have two choices. You can make transmission lines in the backend of line (BEOL) SiO2 and metal layers. The SiO2 dielectric layers are thin, which means high metal losses. Or you can send your wafers to a third party for post-processing to put a lower dielectric metal system on top of it, such as benzo-cyclo-butene (BCB) and gold.

Every time the upper frequency of SiGe extended, the breakdown voltage is reduced. Some of that stuff has to operate at 1.0 volts, which means forget about all but the most girly-man of power amps.

Here is some info from IBM on their SiGe processes which continue to evolve. Notice they don't tell you the operating voltage continues to drop with frequency...

IBM's SiGe HBT BiCMOS Technology Generations:
- 1st Generation (IBM 5HP – 50 GHz HBT + 0.35µm CMOS)

- 2nd Generation (IBM 6HP – 50 GHz HBT + 0.25µm CMOS)

- 3rd Generation (IBM 7HP – 120 GHz HBT + 0.18µm CMOS)

- 4th Generation (IBM 8T – 200 GHz HBT)

- 5th Generation (IBM 9T – 350 GHz HBT)

Examples:

IBM, Motorola

Gallium nitride (GaN)

This is the future of microwave power amps, GaAs has exceeded its half-life, you can quote us on that. More expensive in terms of dollars per die, GaN offers a path to much higher power densities and therefore cheaper dollars per Watt. Breakdown voltages of 100 Volts are possible, soon you will be able to buy 48 volt solid state power amps at X-band! GaN is still a relatively immature process, reliability has been a HUGE problem that is just being overcome. Ancillary stuff like higher-voltage capacitors and resistors, and backside processes need to be redeveloped at MMIC foundries in order to participate in this new technology.

DARPA is pumping millions of taxpayer dollars into GaN so that the US will maintain technological superiority in military programs for the next decade or two. The big DARPA program is called WBGS-II (for wide bandgap semiconductor), and the three teams are TriQuint/Lockheed, Raytheon/Cree and Northrop Grumman. No further discussion will appear here, the data is ITAR restricted!

Substrates for GaN are either silicon carbide, sapphire, or silicon (Nitronix uses this approach). "Native" GaN wafers are impractical, so a lot of expensive alchemy is needed to align the GaN crystal onto mismatched substrates. Four-inch SiC substrates are just becoming available, for GaN-on-silicon, four inch wafers are also available.

Sic is an excellent heat sink, and GaN can operate up to to greater than 150C channel temperature. Below 2 GHz, expect to see GaN used in base station applications, competing with silicon carbide technology. Higher frequency GaN products will be fielded by the military, HRL reports power amplifiers even up at millimeterwave!

Silicon is not such a great heat sink as silicon carbide (40 versus 350 W/m-K), so lower-cost of GaN on-silicon-may be outweighed by the ability to dissipate higher power (and thereby achieve greater power density) on SiC Normally, silicon's conductivity makes it lossy as an RF substrate, Nitronix could fix that using high-resistivity silicon (click here to learn more about microstrip loss due to substrate conductivity). Maybe that is why Nitronix only offers discrete FETs, not MMICs. Eventually the two technologies may find their own niches, the GaN market will be huge.

Examples:

TriQuint, Eudyna, HRL, Raytheon.
Cree for GaN/silicon carbide substrates, and soon they will have their own MMIC line.
Nitronix for GaN-on-silicon substrates and GaN-on-silicon discrete transistors.

Antimonide-based compound semiconductors

At the other end of the power spectrum is ABCs Here's a technology that can operate at only one tenth of a volt! It is possible to create low noise amps that dissipate only one milliwatt using ABCs The market or these? Space-based arrays, where power is limited to solar cells, and the received power from earth is pretty-well attenuated to next to nothing. Don't look for ABCs applications where high linearity is a priority.

Examples:

Teledyne (was Rockwell Scientific)