Technical Briefing PRFI Closes the Gap between Measured ... · were developed using ADS from...

Technical Briefing

PRFI Ltd.The Plextek Building, London Road, Great Chesterford,Saffron Walden, CB10 1NY, UK

T: +44 (0) 1799 796464 E: [email protected] W: www.prfi.com

Custom designed MMICs allowthe desired functionality andperformance to be optimisedwhilst minimising size andpotentially reducing productioncosts. Whilst the upsides areclearly attractive, the con-sequences of a custom designedMMIC failing to meet therequired performance criteria canbe costly. As well as the expense ofthe re-design and a secondfabrication run, the detrimentalimpact on project timescales canbe even more costly. First timeright design success is thereforehighly sought after and it is theability to consistently andaccurately predict performanceprior to manufacture that allowsthis.

Introduction

This white paper reviews some recentMMIC designs carried out by PRFIthat demonstrate how highly accurateMMIC performance predictions can beachived. Close agreement betweenmeasured and modelled performanceis exhibited across a wide range ofcircuit functions, operational fre-quencies and process nodes and alldesigns were first pass successes.Insight is also given into the designmethodology associated with thedevelopment of these MMICs, inparticular detailed EM simulation isessential to achieve such closeagreement. All of the designs presentedwere developed using ADS fromKeysight.

The following design examples arepresented:

● 3rd Order Elliptical High PassFilter for the 39GHz 5G Band

● 6 to 18GHz Single Stage FeedbackAmplifier

● Power Amplifier MMIC for the26GHz 5G Pioneer Band

● 6 to 23GHz Double Balanced Mixer

● 20 to 30GHz SPDT Switch MMICfor 24GHz, 26GHz and 28GHz 5Gbands

● SMT packaged mm-wave MMICs

Sheet Code 0618

PRFI Closes theGap betweenMeasured andModelled MMIC

Performance



6 to 18GHz Single Stage FeedbackAmplifier

The single stage feedback amplifiershown in Figure 3 is, like the filter, asub-circuit from a larger multi-func-tion MMIC. This is a wideband 6 to18GHz gain block designed on a WINSemiconductor 0.25µm GaAs PHEMTprocess. It uses shunt resistive feed-back to cover the full 6 to 18GHz bandwith a flat gain versus frequencyresponse and good input and outputreturn losses.

The measured versus modelled s-pa-rameters are plotted in Figure 4 below.Measured results were made RFOWand are shown in blue with simulatedperformance in red. Agreementbetween measured and modelled per-formance is again excellent.

effort being made to minimize layoutparasitics.

The next stage of the design process isthe most critical and time consumingand must be completed with great carein order to obtain good measured tomodelled performance. The layoutmust be EM simulated, step by stepwith careful optimisation of the designand layout at each step to maintainoptimum performance. Determinationof the most appropriate mesh densityis a compromise between simulationspeed and accuracy. In the case of thisfilter, the input and output 50Ω routingtrack were also included in this proc-ess. They link to adjacent blocks in themulti-function MMIC and have a sig-nificant impact on the performance ofthe filter.

Figure 2 compares the RFOW meas-ured performance of the filter sub-circuit to the simulated. The solidtraces are the measured response andthe dashed traces the simulated. Excel-lent agreement between measured andmodelled is demonstrated to beyond40GHz.

3rd Order Elliptical High Pass Filterfor the 39GHz 5G Band

This filter was designed as part of amulti-function 5G MMIC addressingthe 37GHz (37 – 38.6GHz) and 39GHz(38.6 – 40GHz) 5G bands. Its purposewas to provide rejection below bandwith low insertion loss across the fullRF operating band. It was fabricatedon one of WIN Semiconductors’0.15µm InGaAs pHEMT processes.The filter was realised as an independ-ently testable sub-circuit, as shown inthe photograph of Figure 1.

The design process commences withthe realisation of a simple elliptic filterusing ideal capacitors and inductors(Ls and Cs). The capacitors are thenreplaced by Metal Insulator Metal(MIM) capacitors from the foundryPDK (Process Design Kit) and thefilter is re-optimised. This stage isnormally straightforward and hasminimal impact on performance. Theinductors are then replaced with highimpedance transmission lines, againusing the foundry PDK models. Thefilter is again re-optimised. Layout ofthe filter is now undertaken with every

Figure 1: Photograph of 3rd order elliptical highpass filter test cell

Figure 2: Comparison of measured to simulated



Power Amplifier MMIC for the26GHz 5G Pioneer Band

The EU’s Radio Spectrum PolicyGroup (RSPG) recommended the26GHz (24.25 to 27.5GHz) band as thePioneer Band for mm-wave 5G inEurope. Figure 5 shows a photographof a Power Amplifier (PA) MMICdesigned by PRFI for 5G applicationsin this band. It was fabricated on oneof WIN Semiconductors’ 0.15µmInGaAs pHEMT processes and wasoptimised for best efficiency whenoperating backed-off from compres-

sion to maintain modulation fidelity.The die size is 3.5mm x 1.2mm withscope for reduction to 3.0mm x 1.2mmusing a production mask set.

Figure 6 shows measured to simulateds-parameters, the quiescent bias pointis 210mA from 6V. Measured perform-ance is in blue and simulated in redshowing the measured gain response(S21) virtually overlaying the simu-lated. The S11 is below -18dB acrossthe full 26GHz 5G band. The outputwas matched for best efficiency whenoperating backed-off from compres-

sion but the S22 is still a very respect-able -12dB worst case across the fullPioneer Band.

Figure 7 shows the RFOW measuredand simulated output power and poweradded efficiency (PAE) against back-off from P-1dB at 26GHz. The simu-lated traces use dashed lines and themeasured use solid lines. The outputpower at P-1dB is 26dBm with a PAEof 30%. The measured to simulatedPAE is in very good agreement acrossa wide range of back-off.

Figure 3: Die photograph of the 6 to 18GHzsingle stage amplifier test cell

Figure 4: Comparison of measured to simulateds-parameters for the 6 to 18GHz gain block

Figure 5: Die photograph of a 26GHz 5G PA MMIC



6 to 23GHz Double Balanced Mixer

Other areas in which PRFI has signifi-cant design expertise include MMICmixers. One such design is the 6 to23GHz double balanced mixer shownbelow in Figure 8. This was designedon Qorvo’s 0.13µm PHEMT process,TQP13. It has an RF and LO frequencyrange of 6 – 23GHz and can operatewith IF frequencies in the range of DC- 4GHz. Typical conversion loss isaround 7dB with an IIP3 of 22dBm.The LO to RF rejection is excellent ataround 45dB.

Figure 9, shows conversion gain whenthe mixer is configured as a downconverter with high side RF, and an LOinput power of +15dBm. The measuredconversion gain across the RF inputrange of 6 to 23GHz is within 1dB ofsimulated with an average value of-7dB. The input referred third orderintercept point (IIP3) for the mixer inthis configuration is shown in Figure10; the LO input power was again+15dBm and the tone spacing 10MHz.The average IIP3 across the band isaround 22dBm for both measured andmodelled and for a given RF frequencythe measured IIP3 is typically within2dB of the modelled.

Figure 6: 26GHz 5G PA RFOW measured andsimulated small signal response

Figure 7: 26GHz 5G PA RFOW measured andsimulated power and PAE at 26GHz

Figure 8: Die photograph of 6 to23GHz double balanced mixer



20 to 30GHz SPDT Switch MMICfor 5G

The next MMIC to be reviewed is asingle pole double throw (SPDT)switch that operates with low insertionloss and high isolation across the 5Gbands at 24GHz, 26GHz and 28GHz.It was designed on WIN Semiconduc-tors’ 3µm PIN diode process. The diemeasures 3mm x 1mm and a photo-graph is shown below in Figure 11. TheDC control pads are on the top-side,the common RF port is on the bottomside and the switched RF ports are onthe left and right hand sides.

The switch was measured RFOW withthe left hand side RF path enabled andthe right hand side path disabled.Applying a negative voltage to a path’scontrol pin enables that path (turningthe diodes “off” and setting it into itslow loss state); applying +3V disablesthe switch path (turning the diodes“on” and putting that path into its highloss, isolating, state). The on-stateswitch draws a total of 20mA from the+3V supply. The negative bias on theoff-state arm determines the powerhandling of the switch. A negative biasof -15V allows operation to power

levels of +33dBm without significantcompression.

A plot showing the measured andmodelled S-parameters for the enabledpath is given below in Figure 12showing very low insertion loss (~0.66dB) and excellent agreementbetween measured and modelled. Acomparison of the measured and mod-elled disabled path isolation is givenin Figure 13 demonstrating a very highisolation of over 45dB across the 20 to30GHz operating band.

Figure 9: Comparison of measured to simulatedconversion gain of 6 to 23GHz double balanced

mixer.

Figure 10: Comparison of measured tosimulated IIP3 of 6 to 23GHz double balanced

mixer.

Figure 11: Die photograph of20 to 30GHz 5G PIN diode switch



Figure 12: 28GHz 5G PIN switch measured andsimulated, enabled path s-parameters

Figure 13: 28GHz 5G PIN switch measured andsimulated, disabled path isolation

SMT packaged mm-wave MMICs

The packaging of MMICs into lowcost SMT packages eases handling andassembly and is therefore the preferredformat for most volume applications.At mm-wave frequencies it is essentialthat co-design and simulation of thepackage and MMIC is undertaken asan integral part of the MMIC develop-ment process if optimum performanceis to be obtained.

PRFI has experience of developingmm-wave MMICs using a range ofSMT packaging approaches includingover-moulded plastic, air-cavityplastic and custom laminate. Withplastic packaging a custom leadframedesign is normally used to obtain bestRF performance. The laminateapproach is inherently a customdesign. The transition between theMMIC bond pad and the PCB onwhich the packaged part is mountedmust be modelled and optimisedduring the design process. The effectsof the bondpads, bondwires, PCB and

package leadframe must all beaccounted for and in the case of over-moulded plastic the moulding com-pound itself must be incorporated intothe EM simulation of the MMIC.

Figure 14 is a photograph of a plasticpackaged PA MMIC for the 28GHz 5Gband (27.5 – 28.35GHz) mounted onan evaluation PCB. It was designed ona 0.15µm E-mode pHEMT processfrom WIN Semiconductors and ishoused in a 4mm x 4mm over-moulded20 pin QFN package. A custom lead-frame was designed as part of thedevelopment process to optimise theRF performance of the packaged part.The PA includes an on-chip tempera-ture compensated power detector andoffers a gain of 20dB with a P-1dB of26dBm and a PAE of 30%. The designwas optimised for best PAE whenoperating with IMD3 products at-35dBc (which was around 7dB back-off). At this operating point a PAE ofaround 7 to 9% was demonstrated.

The measured to modelled perform-ance of a packaged PA on the evalua-tion PCB is plotted in Figure 15. ATRL calibration tile was used to refer-ence the measured performance to theports of the package on the PCB.

A custom laminate package was devel-oped for the 26GHz 5G PA MMICfeatured earlier in this paper. Thepackage was a QFN format and housedtwo MMICs to provide a dual channelcomponent as depicted in the photo-graph of the evaluation PCB shown inFigure 16.

A comparison of the RFOW measuredperformance of the PA die to the meas-ured performance of the dual channelpackaged component on an evaluationPCB is shown in Figure 17. Bothchannels of the packaged part areplotted as the solid traces; the dashedtraces are the RFOW measured per-formance. The difference between theRFOW and packaged part perform-ance is modest demonstrating thequality of the RF package design.



Conclusion

This white paper has given an over-view of a proven approach to accu-rately simulating custom MMICs. Across-section of recent design workcarried out by Plextek RFI has beenpresented, which demonstrates con-sistent first pass success through closemeasured and modelled agreement.Such agreement was presented for a

Figure 14: 28GHz 5G PA in 4mm x 4mm overmoulded plastic QFN

Figure 15: Plastic packaged 28GHz PA for 5G,measured to simulated performance

Figure 16: Dual channel 26GHz 5G PA in customlaminate package

Figure 17: Measured s-parameters of two chan-nels of a packaged PA compared to RFOW

range of circuit functions, operationalfrequencies and process nodes.

By cultivating a company culture ofright first time design success, PlextekRFI has enabled its clients to reducetheir product development timescales,cost and risk.

Technical Briefing PRFI Closes the Gap between Measured ... · were developed using ADS from...

Documents

Transcript of Technical Briefing PRFI Closes the Gap between Measured ... · were developed using ADS from...