Jul/Aug 2005 45

Silvano Ricci, IØLVAVia Crocetta, 4000010 S. Polo dei CavalieriRome, Italysilvano.ricci@transport.alstom.comi0lva@katamail.com

A 47-GHz LNA

By Silvano Ricci, IØLVAand

Daniele Moretti, IWØFGR

MMIC Experiences on the 6-millimeter Band

Daniele Moretti, IWØFGRCirc Cornelia 293 pal 5Rome, Italydanielemoretti@tiscali.it

Daniele, IWØFGR, and I thoughtthat we needed to improve ourstations by increasing the re-

ceiver and transmitter performancewith an increase in output power and areduction in receiver NF. We reachedthis conclusion after some intense ac-tivity on the 10-GHz and 24-GHz bandsand a shy attempt on the 47-GHz bandwith no contacts and with just twotransverters based on a sub-harmonicmixer design.

Tackling such high frequencies is

not easy, and difficulties increase ex-ponentially for the following reasons:

• Lack of suitable materials and com-ponents

• Lack of knowledge of the involvedtechnologies, which differ greatlyfrom those used by the majority ofoperators (on HF)

• Lack of equipment necessary forcalibration and measurement

We needed to start by looking for thenecessary active devices available on

the market, which could be used by ushams. We found that as the frequencyincreases, the availability of discrete ac-tive devices such as transistors (in bothpackaged and die forms) decreases. In-stead, MMICs (Microwave MonolithicIntegrate Circuits) are becoming moreand more commonplace, as they arein the lower-frequency bands. Forexample, the ERA and MAR MMICsfrom Mini-Circuits Labs. Yet, there arecertain difficult problems that need tobe overcome to succeed in the millime-ter-wave bands. The products presenton the market have the following draw-backs for use in ham bands :

• The working frequency ranges forwhich these devices are optimizedare about 10-20% of the center fre-quency and cover (rightly) the fre-

Ricci.pmd 6/2/2005, 9:45 AM45

46 Jul/Aug 2005

quency bands of the telecom andmilitary markets. The 10-GHz and24-GHz ham bands are the most for-tunate, but it is not so for the 6-mmband (47 GHz) or higher frequencies(77 GHz).

The MMICs on the market that canguarantee interesting performance areall in the die form, which means wehave no choice but to tackle new tech-nologies and their specific requirements(wire bonding, epoxy attach, handlingand storage of die MMICs).

Before going any further with ourexperiences and realizations it is im-portant to explain some of the techni-cal problems and technologies that wehad to face.

Handling of MMICsMMIC Storage

Die MMICs are usually shipped inspecial containers, such as GEL-PAKtrays (visit www.gelpak.com) orWafflepack trays (visit www.ictray.com). These should be stored in a tem-perature- and humidity-controlled area,preferably under a dry-nitrogen flux.The latter aspect is important in GaAsMMICs if their surface is not passivatedwith NSi4, since hydrogen moleculesthat are adsorbed into GaAs tend tochange its electrical characteristics andmodify the devices’ functional charac-teristics.

GEL-PAK is in any case not recom-mended for long-term storage (typi-cally greater than 1 year) due to pos-

sible chemical interactions betweengel and the backside metallization ofthe die which can bring problems inattaching the device. In case it is nec-essary to store chips for a long time, itis better to have them delivered inWafflepack trays.

All of this is very complicated forus hams, and for the majority of uspractically impossible.

“It’s so small! How do I pick it up?”Well the answer is easy: A pair of cleanstainless tweezers are sufficient. Takecare to pick up the die in a correct way,to avoid chipping the upper edge of thedie (see Fig 1). We recommend buyingtweezers from Dumont (visit www.etweezers.com/) or Fontax.

ESDHigh-frequency GaAs devices are,

for obvious reasons, usually not de-signed with on-chip ESD protectioncircuitry. Due to their reduced geom-etry sizes, MMICs are very sensitiveto ESD degradation or failure. It isnecessary to observe the same rulesused for Class 0 silicon devices.

Assemblies with die MMICsDie Attachment—General Guidelines

There are two methods for attach-ing a die to a substrate or a metal base(see Fig 1A). The choice is generallydetermined mainly by the devices’thermal-dissipation requirements,though attaching the die to a metal

base is the best solution to simulta-neously achieve a good RF ground andthermal cooling. The two methods areconductive die attachment and eutec-tic die attachment.

The general guidelines are:

• Low-power devices can be attachedwith a silver-loaded epoxy.

• Medium-power devices can be at-tached with a silver-loaded organicadhesive (30-60W/mK, milli-kelvins) or epoxy (2-3W/mK), butthis should be limited to devicesmounted in low-temperature envi-ronments.

The recommended method is suit-able for both medium and high-power devices. They should be at-tached with eutectic solder, such as(80/20 Au/Sn).

It is possible to attach low-powerdevices directly onto substrate mate-rials (see Fig 1B), but take care toprovide a low RF ground inductanceunder the die chip and avoid non-TEMpropagation modes or spurious oscil-lations will result. This can be accom-plished by including via-holes thoughthe substrate (see drawings).

Epoxy Die AttachmentEpoxy die attachment is the stan-

dard industrial method. It has lowerproduction costs with good reliabilityof the finished product. From a physi-cal point of view, epoxy attachment isbased on Van der Waals interactionrather than atomic or molecular in-

Fig 1—Die installation on metal (A) or substrate (B).

Ricci.pmd 6/2/2005, 9:45 AM46

Jul/Aug 2005 47

Fig 4—At A, divergent wires connect to the microstripline as noted in the text. At B, a photo of properly mounted MMICS as describedin the text.

Fig 2—A MMIC with external parts mounted on a circular substrate. Fig 3—Thermocompression wedge bonding.

(A) (B)

teraction. Epoxy is therefore an adhe-sive paste composed of two compo-nents, silver or gold grains filled ep-oxy and a hardener, mixed togetherand deposited in liquid form on thesurfaces which are to be attached. Theepoxy Curing times depend on tem-perature. It can be done at room tem-perature, but it is generally better todo it in a ventilated oven at, for ex-ample, 90°C and will take about anhour and a half.

Conductive epoxy can be bought as:

• One-component epoxy (hardener isshipped premixed with epoxy). Thisis a ready-to-use product, but to

slow polymerization, it must bestored at low temperatures (–40°C),but even in these conditions, itsaverage useful life (pot-life) is ap-proximately 6 months;

• Two-component epoxy, which must bemixed prior to use. The mixing pro-cedure and ratios are indicated bythe manufacturer, and they must beperformed in a clean, dry environ-ment. The unmixed components canbe stored at room temperature, withan average storage lifetime of oneyear.

To achieve a good attachment, it isimperative to clean the dispensing

equipment and metal-base chip carri-ers (metal or substrate) with isopro-pyl alcohol to eliminate possible con-taminants. The epoxy can be appliedmanually, with a needle.

It is very important to minimize thethickness of the epoxy layer. First, tokeep thermal resistances low. (Takecare to avoid air bubbles and gaps!)Second, to avoid short circuits causedby the conductive epoxy overflowingthe top of the chip or flowing betweenthe chip, substrate and RF lines.

Learn to make the epoxy drop justa little larger than the chip lead, re-sulting in a narrow fillet around the

Ricci.pmd 6/2/2005, 9:46 AM47

48 Jul/Aug 2005

contact! Position the chip directly atits final position and press it slightly,making sure not to damage the topsurface of the die. Chips have delicatecomponents, such as air-bridges, andspiral inductors, which are damagedvery easily.

Here are some references for epoxies:

• EPO-TEK H20E from EPOXY TECH-NOLOGY; www.epotek.com

• ABLEBOND 84-1LMI fromABLESTIK; www.ablestik.com

• DM6030HR from DIEMAT; www.diemat.com

QMI5030 from the former DEXTERELECTRONIC MATERIALS (www.dexelec.com); as of August 2003, Dex-ter was bought by LOCTITE (www.loctite.com).

Eutectic Die AttachWe won’t cover this topic section in

this article. But maybe next time,when we do some high-power MMICassemblies!

BondingClassic (SnPb) tin-lead soldering is

not possible with die MMICs becauseof incompatibility between SnPb andgold (gold is present on the chip padsand microstrip) and because MMIC di-mensions are much smaller (typical di-mensions are 1/2 mm2) than with otherchips. Needless to say, with these verysmall dimensions it is fundamental toobserve everything through a micro-scope! Other bonding materials andtechniques are therefore necessary.

Bond wiresThe bond wires used to connect the

MMIC to the microstrip lines on theexternal substrates are usually puregold, 18 µm to 25 µm in diameter,which have an inductance in the or-der of 0.6 nH/mm to 0.8 nH/mm.MMIC designs account for the induc-tance effect of bonding wires in the

Table 1—Characteristics of Thermocompression Bonding Techniques

Characteristic Wedge-Bonding Ball-BondingFootprint Very small,

1.5 to 2 times wire diameter Large, 3 to 5 times wire diameter

Length Shortest possible Longer, wire starts off vertically

Wire Size Capability ≥ 18 µm ≥ 18 µm

Speed SLOW2 separate alignments necessary;wire needs to be moved exactly under tool end.

FASTomnidirectional movement of tip;wire is fed directly under tool end.

final chip performance, so it is desir-able to strictly adhere to the MMICdesigners recommended mountingspecifications.

The best way to achieve the elec-tric transition from the coplanar-with-ground structure to the microstripstructure is by using two divergentbond wires as shown in the Fig 4A.This can only be conveniently accom-plished with 18 µm-thick wire and bythermosonic wedge-bonding as de-scribed in the next paragraph.

This pair of bond wires must be nolonger than 200 µm in order have anoverall inductance value ranging from0.2 nH to 0.4 nH.

Bonding ProcessesThe two main types of bonding pro-

cesses are:1. Ultrasonic bonding (very popu-

lar in laboratories);2. Thermocompression bonding (the

preferred industrial method) is sub-divided into:

A. Thermocompression wedge bond-ing (see Fig 3)

B. Thermocompression ball bonding

Ultrasonic bonding uses ultrasonicenergy (typically 60-100 kHz) to in-crease the plasticity of metals to bebonded. The ultrasonic system of a wirebonding machine consists of two parts:the ultrasonic generator and the ultra-sonic transducer. Here is the bondingsequence: First, under the applicationof force by the wedge tip, a certainamount of deformation occurs in thelattice structure of the bond wire and/or bond surface. Next is a cleaningphase. Ultrasonic energy (with ampli-tude vibrations of 1-5 µm, much smallerthan the bond-wire diameter) makesthe wire and wedge move together, cre-ating friction at a constant pressure onthe wire and bond interface surface.Shortly, the wire deforms and heats sothat welding occurs. (Welding occurs by

the diffusion of the wire and bond sur-face-lattice dislocations.)

Thermocompression uses a combi-nation of heat and pressure to connectthe wire between the die bond pad andthe microstripline on the external sub-strate. No melting between metals oc-curs as the bond comes about by theinteraction of atomic forces betweenthe wire and the metal pad.

Thermocompression Wedge Bondinguses a hard heated wedge-bonding tipmade of tungsten carbide together witha wire spooler equipped with a wireclamp. The wire is pressed on the bondpad with a controlled force, typically 20to 22 grams. This process requires aprecise alignment of tool force, workstage and tip temperatures.

Thermocompression Ball-Bondinginstead has the wire fed through a cap-illary in the tip, which is heated to ahigh temperature (300-400°C). A hydro-gen flame or spark discharge is pro-duced at the end of the tool to melt theend of the wire and form a small ball.The tool then moves over the bond padof the die and presses vertically (forceis around 30 to 50 grams) the ball onthe pad to realize bonding. The tip ismoved to the new location (externalsubstrate) and wire is again pressed (noball is formed in this case) to achievethe new bond and thereafter cut.

Bonding process control should becompleted to validate the die attachprocess control over time (repeatabil-ity) by pull testing. Pull-testing maybe destructive or non-destructive de-pending on the test method, but weshall not discuss that here.

Substrate Materials forMicrowave

Choice of the appropriate micro-wave substrate material takes intoaccount many factors such as :

• frequency of operation;• cost;• thickness;

Ricci.pmd 6/2/2005, 9:46 AM48

Jul/Aug 2005 49

Table 2—Microwave Circuit Materials

Minimum CopperAvailable Thickness Loss

Substrate Thickness - unplated ε Tan Dimensional(mm) (µm) (µm) @10 GHz @10 GHz Stability

Rogers RO4003 127 ±10 17-70 3.38 ±0.05 0.0027 GoodRogers RO5870 127 Tol. ? 13-35 2.33 ±0.02 0.02 PoorRogers RO5880 127 Tol. ? 13-35 2.20 ±0.02 0.02 PoorTaconic TLY 3 127 Tol. ? 17-35 2.33 ±0.02 <0.02 Fair

• size/Dk;• dimensional stability.

There is a drawback though.Duroid, though its electric propertiesare superior to other materials atthese frequencies, is unfortunately notthe best choice for thermosonic bond-ing . It is a soft material and the tem-peratures that are reached in themicrostrip region tend to detach thecopper strip from the substrate. Aharder material would perform better.

Our experimentsAt 47 GHz, it is convenient to use

Teflon-based copper-laminated dielec-tric materials (ε>> 2.3, with a flash ofgold on the copper of approximately3 µm to avoid oxidization) 5 mils thick(127 mm). This is not only an electricalnecessity (microstrip propagationrequires that the height of the dielec-tric material be many times smallerthan the electric wavelength) but alsoa mechanical necessity. MMICs are usu-ally about 100-127 µm thick, whichmeans that the RF pads of the MMICare at nearly the same height as theRF microstrip lines. This simplifiesbonding, which would otherwise requirea “deep-access” bonding tool. It also en-sures that the bond wires remain asshort as possible, minimizing their to-tal inductance, which turns out to bean electrical requirement, as well.

In this case, a 50 Ω microstrip lineis approximately 350 µm wide, whileon the MMIC the RF pads are usuallyabout 70 µm wide and usually arecoplanar with ground lines.

The MMICs meet the following con-ditions for proper mounting and per-formance (see Fig 4):

• The low-power MMICs were gluedwith conductive epoxy (like H20EEPO-TEK) on top of a conductivemechanical holder.

• There are two 50-Ω microstrip linesfor the input and output, each atleast two electrical wavelengths (inmicrostrip) to allow for tuning. Tun-ing was accomplished by movingshorting bars made of copper (witha gold finish). A gold-loaded conduc-

tive epoxy, such as H81E (from EPO-TEK) can be used, but it is very ex-pensive! The microstrip lines wereglued with a conductive epoxy (suchas H20E) to the mechanical holderor at the bottom of the cavity. It isbetter to obtain some 5 mil 5880 or5870 Duroid (ROGERS Corp.) thathas one side with coated with½-ounce ED copper and the otherside with thick (60 or 90 mil) EDcopper. Vendors can provide their di-electric material in such form uponrequest. This solution is apparentlycostly, but it does not require a me-

chanical holder; it has sufficientmechanical strength.

MMICs are typically powered froma 4 or 5 V supply through the FETdrains and or more variable voltagesto set bias condition of the FET (gates).Typically, this voltage is negative(–5 V). Ensure that the negative biasvoltage is applied before the positivedrain voltage. Otherwise, permanentMMIC damage may result. Refer tothe component datasheet for power-supply details.

Power supplies must be carefully

Table 3—Measured Results for a Single CHA2094B

Parameter Prototype 1 Prototype 2Gain 14.3 dB 13.8 dBNF 5.30 dB 4.78 dB

Fig 5—A schematic of the single-MMIC circuit.

Ricci.pmd 6/2/2005, 9:46 AM49

50 Jul/Aug 2005

filtered. The filters are usually madeof a small inductance (the bond wireforms a RF choke) and an RF-shuntcapacitor mounted (epoxy-glued) di-rectly near the power-supply pads ofthe MMIC. Select appropriate capaci-tors carefully. MIS (metal-insulator-semiconductor) capacitors are verygood capacitors for shunting mm-wavefrequencies. It is usually good practiceto follow the indications given in theMMIC datasheet, since the MMICdesigners account for a recommendedpower supply and bias circuit whenoptimizing design performance.

The MMICs We UsedLast year (2002), we were investi-

gating what the MMIC market couldoffer to hams. The MMIC market issubstantially held by a few companies,such as Filtronic, UMS, Agilent, TRW-Velocium, Mimix, Raytheon andTriQuint, Some of these have internalGaAs MMIC foundries, while othersare mainly MMIC design houses.

Our research resulted in the choiceof two MMIC amplifiers (all data-sheets are available through theInternet).

CHA-2094B from UMS: 20-dBgain, three-stage low-noise amplifier(LNA) with a maximum output powerof 8 dBm and a NF of 0.75 dB.

HMMC-5040 from Agilent: a four-stage medium power amplifier (MPA)to work in the 20-40 GHz band with atypical gain of 22 dB, a 1-dB compres-sion point of 18 dBm and 21 dBmsaturated output power. No further in-formation was available at 47 GHz (S-parameters and so on), but a closelook at the gain curve made us feel itsperformance could be acceptable inour frequency band.

Our PrototypesFig 5 shows the circuit we used for

our prototypes. The positive (drain)power supply is designed around a7805, and the gate negative voltage isdesigned around an ICL7660. The de-vices require three different bias volt-ages (input stages need Vgs1 and e Vgs2,Vgs3 is for the output stage).

The nominal bias current is approxi-mately 20 mA for the output stage and15 mA for each input stage; for a totalpower consumption of 50 mA. The bestNF performance is accomplished with45-mA total power consumption.

We have measured performance ofthe two MMICs in both single and cas-cade arrangements.

The CHA2094b, alone, yields poorerresults (both in NF and in Gain) thanwe expected based on the datasheet.Even in-band (36-40 GHz) measure-

Fig 6—At A, a photo of thetest setup with single andcascade MMICs and atransverter. At B,characteristic curves forthe LNA with CHA2094Band cascaded transceiver.

Table 4

Parameter Prototype NotesGain 26.8 dB 36.1 dB together with the transverter of Fig 6NF 5.10 dBI 90 mA

ments exhibited 5 dB more noise thanon the datasheet. This is not justifiedby the losses of the waveguide-to-microstrip transitions and losses inthe input and output microstrip lines.

Anyway, the results in Table 3 werestill very interesting. The measuredvalues must be considered averagevalues since they depend on tempera-ture, which we didn’t measure.

Table 4 shows the performance forthe two modules connected in cascadeand to the transverter.

The same preamplifiers have beenused in both transmission and recep-tion by means of a WR22 waveguidetransfer relay, and the power measure-ments have given these results:

Input –6.32 dBmOutput 12.19 dBm,

approximately 16.4 mW(near saturation)

The output shall be used to feed apower amplifier we are designing andwe will describe that in our next article.

The output power at the 1-dB com-pression point, even though it varieswith temperature, once stabilized, hasbeen around 10.5 dBm, which is in-deed a good result.

We performed measurements onthe cascade of two HMMC-5040MMICs and the results were a gain>>25 dB; NF ≈ 8 dB; Pout ≈ 15 dBm

(A)

(B)

I0LVA - IW0FGR - EATON 2075 - LNA 47 GHz 2 x

CHA2094B

4

4.5

5

5.5

6

6.5

47084

47085

47086

47087

47088

47089

47090

47091

47092

47093

47094

Freq. MHz

dB

NF

33

33.5

34

34.5

35

35.5

36

36.5

37

dB

Gain

N.F. dB Gain dB QX0507-Ricci06b

Ricci.pmd 6/2/2005, 9:47 AM50

Jul/Aug 2005 51

built by our friend Armando, I3OPW,in brass (with no silver plating or othersurface treatment). Once the filter wastuned, our first measurements with anetwork analyzer showed excessive in-band insertion loss of 5 dB, LO attenu-ation (46,944 MHz) of 50 dB and>60 dB (beyond the sensitivity of ourinstrument) for fIM (46,800MHz). Byincreasing the diameter of the cou-pling hole, we managed to reduce in-band insertion loss to 2.6 dB (maxi-mum) without too much degradationof the out-of-band attenuation levels(>30 dB for LO and >50 dB for fIM),which is acceptable. These measure-ments are shown in Fig 7. We believethat better results could be obtainedwith surface plating of the filter ca-pable of enhancing surface conductiv-ity. Fig 8 is a drawing of the filter.

Special Thanks to…We send a special thanks to our

friends who have given us materialand moral support.

We are now operational and look-ing to correspond with other hams todo trials or contests. Our local DX fieldtrials at 47 GHz have produced verygood results, but there are still manyimprovements to be done. A specialthanks to M.E.D.S. sas for the assem-bly of our first prototypes and techni-cal/technological skills and advice.

Daniele Moretti, IWØFGR, has a de-gree in Electronics Engineering fromthe University of Rome. He is workingas an RF Engineer. He has been a li-censed amateur since 1994 with a spe-cial interest in the microwave bands.

Silvano Ricci started his radio inter-est at the age of 14, as an SWL. He waslicensed in 1969 as IØLVA. He holds acertificate in electronics and telecom-munications. In 1970, he was at the topof the class in the electronics at thetransmission school in Naples. He hasa special interest in VHF-UHF and themicrowave bands from 5.7 GHz to145 GHz. He has published articlesin Radio Revista and DUBUS. He onthe roll of honor of the AssociazioneRadioamatori Italiani.

Fig 7—Response curves for the image-rejection filter.

The Test BenchThe following measuring instru-

ments were used:Noise Figure Meter Eaton 2075BNoise Figure Meter HP 8970ANoise Source HP 346ANoise Source HP Q347BPower Meter HP 346APower Meter HP 435BPower Sensor HP8487A

Fig 8—Mechanical details of the image-rejection filter.

and 17 dBm in saturation. The dcpower requirement is about 550 mA.

The Image-Rejection FilterReceiver input noise forced us to

use an image-rejection filter in our re-

ceiving chain. Here’s a brief introduc-tion and description of our approachwith this type of filter in the 6-mmband. The filter we used is a doublycoupled resonant-cavity filter de-signed in a circular waveguide and

Ricci.pmd 6/2/2005, 9:47 AM51