cmos receivers for active and passive mm-wave...

IEEE Communications Magazine • October 2011190 0163-6804/11/$25.00 © 2011 IEEE

AN OVERVIEW OF MM-WAVEIMAGING SYSTEMS

Imaging at millimeter-wave (mm-wave) frequen-cies has recently gained interest for applicationsranging from security screening to bio-imagingand spectroscopy. Unlike other regions of theelectromagnetic spectrum, mm-wave offers theunique ability to penetrate clothing and other tex-tile materials, enabling new approaches forweapon and contraband detection. Additionally,the transmission and reflective responses of manymaterials have distinct features within the mmand sub-mm wave bands, allowing for the possi-bility of remote material identification [1]. Suchsystems could potentially detect explosives, poi-sonous materials, and other chemical threats fromseveral meters away. While the basic buildingblocks of mm-wave imaging systems are similar tothose used in wireless communications (transmit-ters, receivers, detectors and antennas, etc.), theconfigurations and requirements for mm-waveimaging are quite different. Unlike mm-wavecommunications that wirelessly transmit andreceive modulated signal/data through an mm-wave carrier, mm-wave imagers detect features ofan object based on its radiation, reflection orabsorption properties in the mm-wave spectrum.

Traditionally mm-wave imagers were imple-mented in III-V compound semiconductor tech-nologies for their superior noise andhigh-frequency performance when comparedwith silicon counterparts [2–4]. Although a III-Vbased imaging system does offer higher perfor-mance, it also has its own limitations includingexcessive pixel power consumption, limited inte-gration in forming two-dimensional receiverarrays, pixel readout circuits, and digital signal

processing (DSP) electronics. In response tomany emerging applications of mm-wave imag-ing and the limitations of III-V technology, thesilicon community has proposed severalapproaches to implement mm-wave imaging sys-tems in complementary metal oxide semiconduc-tor (CMOS) technology based on both simpledetector [5] and pre-amplified detectorapproaches [6–8]. Each of these approaches pro-vides solutions to address the fundamental chal-lenges of mm-wave imaging systems in operationmodes, detection time/range, power consump-tion, and portability.

There are several challenges that must beovercome to realize mm-wave imaging systemsin CMOS. First, illumination source (i.e., trans-mitter) power in active imaging remains quitelimited in silicon, so high receiver sensitivity isextremely crucial to balance the link budget forlonger-range or stand-off imaging applications.Second, detection or receiving at higher frequen-cies (> 100 GHz) can lead to higher image reso-lution based on wavelength diffractionconsiderations of available optics. Limited byintrinsic device gain near cutoff frequencies (fmaxand ft), the CMOS low noise amplifier’s (LNA’s)gain is relatively low above 50 GHz, whichbecomes a major drawback to high resolutionimaging system design. Third, wide front-endbandwidth and high detector responsivity areessential to achieve high quality passive mm-wave imaging due to the extremely weak receivedsignals. Such performance is again more difficultto be obtained in CMOS due to its limited gain-bandwidth product versus that of III-V devices.The noise performance of CMOS detector is fur-ther exacerbated by it’s stronger presence offlicker (or 1/f) noise, which requires additionalcircuitry to calibrate and compensate.

Despite all of its challenges, CMOS mm-waveimaging is still highly desirable due to its com-patibility with mainstream integrated circuit (IC)technology and its great promise to offer a low-power, fully integrated, and compact system-on-a-chip (SoC) solution for mm-wave imagingsystems.

IMAGING SYSTEM CONFIGURATIONSThe detection of objects based on mm-waveradiation or reflection/absorption can be dividedinto three separate approaches [1] as shown in

ABSTRACT

This article discusses recent advances inCMOS-based mm-wave receivers for both passiveand active mm-wave applications. System levelconfigurations including scanning and array con-figurations are discussed as well as the advan-tages of pre-amplification vs. simple detection foractive and passive imaging applications. Finally,several recent mm-wave receviers implementedin CMOS technology for both passive and activeimaging applications are reviewed in detail.

INTEGRATED CIRCUITS FOR COMMUNICATIONS

Adrian Tang, Qun Jane Gu and Mau-Chung Frank Chang

CMOS Receivers for Active and Passive mm-Wave Imaging

TANG LAYOUT 9/22/11 6:35 PM 90

IEEE Communications Magazine • October 2011 191

Fig. 1, each with a unique set of requirementsplaced on the mm-wave detector circuitry and itssupporting optics.

The first approach, commonly named trans-missive imaging and depicted in Fig. 1a, is rela-tively straightforward when compared to otherapproaches. A target is placed between a trans-mitter and receiver. The attenuation or blockingof the target at mm-wave frequencies is thenmeasured at multiple points in space to con-struct an image. While this is the simplest typeof imaging, it inherits a major disadvantage ofrequiring the system to surround the target forremote security screening and contraband detec-tion.

The second approach, shown in Fig. 1b, is tofocus the source of mm-wave illumination (trans-mitter) into a narrow beam, then reflect the signaloff the surface of a remote target and return itback to a receiver. This approach is commonlycalled backscatter or reflective imaging. The reflec-tion coefficient information at different points inspace is utilized to create an image. While this isconceptually simple, the signal must travel thepath twice (forward and backward) in free-spaceand makes the link budget more difficult.

The third approach, commonly called “pas-sive imaging,” is shown in Fig. 1c. It constructsimaging by detecting subject’s own blackbodyradiation, without the need of mm-wave radia-tion sources or any compliance to FCC regula-tions. That can simplify the system and facilitateits wide applications. However, passive imagingis considered to be the most difficult approach inmm-wave imaging systems because it must detectthe emitted thermal noise kT from the target,where k represents the Boltzmann constant andT represents the absolute temperature measuredby Kelvin (or K). While the concept seems rela-tively simple as compared with communicationlinks, it is actually quite challenging as the tem-perature variation across a typical scene (con-trast ratio) may vary only by a few K. Sincescene content varies by such small temperaturesa passive mm-wave imager requires <1K in tem-perature resolution to capture usable images forobject detection. One major challenge is thatthis level of thermal resolution often corre-sponds to several orders of magnitude lowerthan the receiver’s input referred noise. For thisreason, techniques such as double-sampling andlong windows of integration must be implement-ed to process the receiver noise to levels wherethe target small temperature difference becomesdetectable. One extremely important applicationof passive mm-wave imaging is the radio-tele-scope used for astronomical study, where thenoise integration window typically lasts severalweeks per frame [1].

The scanning or capturing configuration of animager is another important issue. Regardless ofimager type (transmissive, reflective or passive),measurements must be performed at multiplepoints in space for image construction. To obtainsufficiently high image quality for typical securityor weapon detection applications, the requiredmeasurement points may approach a very largenumber (> 105 pixels), which would seriouslychallenge the system’s power consumption andform factor as indicated in the later sections.

For active imaging, it is possible to creategeometries where a single transmitter providesmm-wave radiation that can illuminate all pointsrequired to form the image simultaneously. Thereceiver used to construct the image, however,will still need to be positioned throughout eachpoint in space. The obvious solution to ease theneed of high pixel count is time-sharing thesame pixel to perform measurements over multi-ple-points. Based on this time-sharing concept,there are three possible configurations that anmm-wave imaging system may choose.

The first and most obvious configuration isshown in Fig. 2a using a single receiver for eitherpassive or active imaging and letting the targetbe scanned through optical or beam-formingmeans. While this is the simplest receiver config-

FFiigguurree 11.. a) Transmissive mm-wave imaging system; b) reflective (backscat-tering) mm-wave imaging system; c) passive mm-wave imaging system.

Path loss

Path loss2

Path loss1

(a)

(b)

(c)

Illuminationsource

Illuminationsource

Imagingreceiver

Imagingreceiveror detector

Target

Insertion loss

Target

Reflection loss

Target

Imagingreceiveror detector

KTradiation

TANG LAYOUT 9/22/11 6:35 PM Page 191

IEEE Communications Magazine • October 2011192

uration, it presents several major challenges. Thefirst is to attain a sufficiently small beam-widthto facilitate the scanning mechanism so that theimager’s resolution can be maintained. The sec-ond is to meet the finite frame time requirementfor video applications. Since the frame timemust be divided among all constituent pixels asshown in Fig. 2 right, it consequently imposesstrict limits on the permissible sampling time ofthe receiver. This requirement on sampling timebecomes exceptionally difficult in passive imag-ing where long integration times, on the order of1ms, are required for each pixel. An additionalconsideration is that the optical or beam-formingsystem becomes very complicated because itmust move in two dimensions as opposed to one.For an optical system, this also hinders reliabilityas it leads to large numbers of moving parts andrelatively high power consumption as the opticsare in continual motion.

The second option shown in Fig. 2b is toexploit a line or linear scanner, where manyreceivers are constructed on one chip in a one-dimensional array and scanned in the perpendic-ular direction. While a line scanner is considereda balance between the other two configurations,

it still incurs the problems of both. Pixel sampletime is still relatively short for passive imaging,and the system must still incorporate movingoptics or beam-forming with extremely fineangular resolution. Using multiple receivers alsointroduces new problems. The power and area ofthe receiver is inflated by “n,” proportional tothe number of pixels in the linear array. Thelength of the array must fit within the diameterof the wafer to facilitate fabrication on a singlesubstrate and hence further limits the pixel area.While in some systems multiple sections aremechanically combined into one array, thisincreases cost and complexity and defeats themajor advantages of using CMOS. Other issuesmust be considered, similar to those found inmulti-channel communication transceivers,including the antenna coupling, power distribu-tion, and the local oscillator (LO) distribution.

The third and final option is to implementpixels into a two-dimensional array of receiverson one chip (often called a focal plane array),which greatly relaxes pixel’s sampling timerequirement and with the extra advantage ofusing non-moving optics. Despite those benefits,it creates difficulty in power and area budgets as

FFiigguurree 22.. Scan configurations for mm-wave imaging systems and their timing requirements: a) single pixel scanning imager; b) linearscanning imager; c) full focal-plane array imager.

Single pixelscanning imager

Linearscanning imager

Focal planearray imager

2 axisscanningoptics

1 axisscanningoptics

Fixedoptics

Frame rate (Hz)105

100n

10u

Max

imum

pix

el s

ampl

e ti

me

(s)

1u

100u

1m

10m

100m

1

15 20 25 30 35 40

Target(a)

(b)

(c)

Target

Target

Full-array n X n

50 linear

100 linear

500 linear

50 single pixel

100 single pixel

500 single pixel



the pixel size and power consumption are inflat-ed by n2. This makes the receiver design particu-larly difficult. For example, if an array wasconstructed with receiver that consumes 100mWof power, a 100 × 100 full array would require akilowatt of power. A 1mm2 receiver implement-ed in a 100 × 100 array would require 10cm ×10cm of chip area. Similar to the linear array,the typical issues of multiple receivers will stillneed to be considered.

SIMPLE AND PRE-AMPLIFIEDDETECTION

In communication systems, detection and recep-tion are terms often used to differentiate thereceived signal-to-noise ratio (SNR), wherereception tends to have a higher SNR. Imagingalso makes a similar distinction and can be parti-tioned into simple detection and pre-amplifieddetection. Figure 3 contains the block diagramsof both a simple detector (left) and a pre-ampli-fied detector (right). While they are similar inconstruction, the order of the blocks is different.

For mm-wave imaging, the distinctionbetween simple detection and pre-amplifieddetection is at which the first gain stage occurs.This is critical as it determines not only the noiseperformance, but the bandwidth and range offrequencies the imager is sensitive to. Amplifiersat mm-wave frequencies are typically narrow-band while nonlinear mixing elements can bequite broadband in nature.

In the simple detector (Fig. 3, left), a nonlin-ear element first translates an incoming signal toa very low frequency (near DC) voltage and sub-sequently amplifies it by a chain of low frequen-cy amplifiers. As the detector has nopre-amplification, the noise equivalent power(NEP) of the detector is typically up to several .Consequently, the detector’s input referred noiseis substantially higher than the target thermalnoise, limiting its usage for passive imagingapplications because of the excessive integrationtime required to overcome the detector noise.Given the high level of noise, simple detectionsensitivity is typically ranged between –30 to –40dBm, which seriously restrains its use in reflec-tive or backscatter systems. Its front-end elementis however non-linear at any frequency and cantherefore be operated at frequencies above thedevice fmax, defined as the frequency at which adevice delivers unity power gain. Nonetheless, asthe frequency increases beyond fmax, the NEPincreases drastically and the responsivity (voltsout per watt in) falls off correspondingly. At fre-quencies beyond 300 GHz, diode based non-lin-ear detectors and self-mixing based lineardetectors [5] are employed to maintain usefulresponsivity and NEP up to 1–2 THz range.

Pre-amplified detectors (Fig. 3, right) behavequite differently from simple detectors as theirfirst stage is made of a low noise amplifier(LNA) allowing suppression of noise from fol-lowing cascaded stages. For this reason, pre-amplified detectors can achieve much betterNEP, as low as 10s of fW/√

——Hz range. As a result,

the pre-amplified detector must operate at fre-quencies well below fmax to ensure sufficient

LNA gain. While a wider bandwidth B is desiredfor the passive imaging to obtain higher receivedthermal power (KTB), a narrower bandwidth isfavored by the active-imaging to increase thereceiver sensitivity and decrease the sourcepower requirements. Typical mm-wave pre-amplified detectors’ sensitivity ranges are from–50 to –60 dBm. Since all passive imaging andmost active mm-wave imaging systems are non-coherent in signal receiving (no phase informa-tion required), they can be constructed by usingan LNA followed by a detector. The detectorsimply detects received radiation power from theLNA output so that the attenuation, reflection,or noise measurements can be carried out atvery low frequency by a high-resolution and low-speed ADC.

CMOS APPROACHES FORACTIVE MM-WAVE IMAGING

Active mm-wave imaging usually works in eithertransmissive or reflective mode. In transmissivemode operation, as an example shown in Fig. 4,a target is placed between a source and a simpledetector or pre-amplified detector. The attenua-tion of the source illumination is then measuredat each point of the perpendicular plane to pro-duce an image. This is by far the easiest in termsof meeting the system link budget as the free-space travel is only in one direction, keeping thepath loss lower than that of the reflective one.This type of system is typically implemented byusing a CW source, a lens in the “optical” pathfor beam focusing and a detector/pre-amplifieddetector array. Typical power needed for theCW source is between 5–10 dBm and the detec-tor/ pre-amplified detector sensitivity is project-ed to be –30 to –40 dBm. With a path-loss of 10dB typically for a 10 cm target distance, this pro-vides an SNR of 30 to 40 dB which is more thansufficient for capturing details of small objects.For example, [5] presents a transmissive modedetector array, operating at 650 GHz in a 0.25μm CMOS process to capture object imageinside an envelope, by using a 0 dBm sourceover a gap distance of 10 cm. The detector itselfis made of a single MOS device to sense the sig-nal by a resistive self-mixing technique.

Reflective or backscatter mode imaging ismuch more difficult than its transmissive coun-terpart as the path loss is doubled (the signalmust first travel from the imager to the target,then reflect back from the target to the imager)and the reflection coefficient of most materials ison the order of –20 dB, which further exacer-bates the link budget. Considering the same illu-mination power of 10 dBm as in the previous

FFiigguurree 33.. Simple detector and pre-amplified detector block diagrams.

Simple detector Pre-amplifieddetector



transmissive case, a path loss of –20 dB for asimilar target distance and a reflection coeffi-cient of –20 dB (typical clothing), the receivedpower at the imager would be only –50 dBm,which is far too low for detection based schemes.Modulation of the source can be used to improvedetector sensitivity by moving the operatingpoint away from the device flicker noise but thesource bandwidth quickly becomes the limitingfactor. Even with modulation, the best detectorsare still limited to around –45dBm sensitivity [6].A typical reflective imaging setup and its testingresults are shown in Fig. 5a. With a –50 dBmreceived power, an imager sensitivity of –60 to–70 dBm is needed for a large enough SNR(about 10 to 20 dB) for image construction.

Although a conventional receiver with anLNA can be used for enhancing backscatteringperformance [8], it becomes very difficult forconstructing an imaging array. Conventionalreceivers consume high-power and area due totheir high stage-counts, large bias currents andinductive loads all necessary to achieve usefulgain above 100 GHz.

In contrast, [9, 10] presented an alternativeapproach by using super-regenerative CMOSreceivers to act as an imaging pixel. Unlike otherreceiver topologies, the super-regenerativereceiver uses a single-stage topology and senseschanges in the startup time of an oscillator toprovide gain at mm-wave frequencies. One majoradvantage of this approach is that it can operateat frequencies much closer to device’s maximumoscillation frequency of fmax, as it does notdirectly rely on device gain for amplification.

Figure 5b illustrates the operation of a digitalregenerative receiver (DRR) with a super-regen-erative front-end but a novel time-encoded digi-tal output scheme. The receiver invented by theauthors functions as the latch is first set by thesystem digital clock edge for engaging the oscil-lator. Once the oscillation envelope rises beyonda prescribed threshold, the envelope detectorwill be agitated to reset the latch and thusquench the rise of oscillation. The DRR princi-ple suggests that the time between the latch setand reset is inversely proportional to the loga-rithmic input power. This DRR imager, under

an illumination source of 183GHz, was testedwith –72 dBm sensitivity and 1.4 GHz signalbandwidth. It consumes only 13.5 mW and occu-pies an area of only 1.31 × 104 um2/pixel. Com-pared with other approaches, it demonstratesorders of magnitude improvement in NEP,responsitivity and power/area consumption.Table 1 compares the performance of the DRRto that of other imaging receivers.

It suggests the DRR can support higher imag-ing resolution with better sensitivity than that ofconventional receivers while also relaxing powerand area budget for full array system applica-tions.

Unlike transmission mode imaging, reflectiveimaging also has additional challenges related tothe nature of reflection at mm-wave frequencies.As shown in Fig 5, many artifacts exist producinga low image quality even though the receiver’sSNR is as high as 20dB. The first artifact is“image speckle” caused by the narrowbandnature of the imager and the comparable sourcewavelength versus the object’s surface featuredimensions. Speckle causes smooth and uniformsurfaces to appear rough and spotted in cap-tured images like the bottle in Fig. 5. Speckleresults from constructive and destructive inter-ference of reflective waves from different surfacedepths. Several approaches are being proposedto reduce speckle with imaging at multiple fre-quencies and averaging received results to allevi-ate surface interference.

The second and more difficult artifact toremove is “clutter,” which is caused by strayreflections from surrounding objects as seen inFig. 5a. The captured image of the tape roll con-tains several artifacts which are not in the origi-nal scene. Since the antenna itself may haveside-lobes, and the optics may also exhibit somereflectivity, it is possible to get radiation returnedthat is unrelated to the target reflection. Theclutter can be addressed by taking images atmultiple angles or through double samplingtechniques to first capture the known systematicclutter and then subtract it from the capturedimage.

The third and most difficult artifact in reflec-tive imaging is the issue of “incidence.” Placing a

FFiigguurree 44.. Transmissive mode imaging and capture from a multi-regenerative receiver (MRR) imager at144 GHz [10].

Speckle results from

constructive and

destructive interfer-

ence of reflective

waves from different

surface depths.

Several approaches

are being proposed

to reduce speckle

with imaging at

multiple frequencies

and averaging

received results to

alleviate surface

interference.



relatively smooth surface at an off-axis angle tothe plane of observation may reflect all thepower away from the receiver or detector andcreate an artificial dark spot. This causes keyfeatures such as the handle of the gun in thetop-right image of Fig. 5 to evade the image cap-ture and create problems for object recognitionin security screening applications. On the otherhand, large bright spots can also be created inthe image, if the incident energy is concentratedby a curved or concave surface on the target tosaturate the receiver. A few solutions have beenproposed to address the incidence problembased on known radar techniques, such as range-gating and multiple angle imaging. One simpletechnique to overcome incidence is to plot thereceived pixel data on a logarithmic scale insteadof a linear scale. While this is easy to implement,it requires much higher SNRs (> 50 dB), whichagain places more challenging requirements on

receiver sensitivity and illumination sourcepower. Other techniques including multi-bandimaging and dual-polarized imaging (in whichtwo antenna polarizations are imaged separate-ly) also show some ability to reduce the highincidence contrast of reflective imaging systems.

CMOS APPROACHES FORPASSIVE MM-WAVE IMAGING

Today’s mm-wave passive imaging systems aremainly based on III-V technologies, such asGaAs or InP HBTs/HEMTs [3, 4], which aregenerally less integrated, thus more sizable andexpensive for array applications. To realize a sin-gle-chip imager with small form factor and lowpower consumption, advanced CMOS technolo-gy would be a better choice. However, the tech-nology comes with higher substrate, metal and

FFiigguurree 55.. a) backscatter imaging setup and images of common objects captured at 183 GHz; b) circuitschematic of 183 GHz DRR receiver [8]; and c) die photo of the 183 GHz DRR receiver in 65 nm CMOS

t

Antenna Osc

CLK

(b) (c)

(a)

CLK

Oscillator ENVDCT

QS R

tQ

t

One simple tech-

nique to overcome

incidence is to plot

the received pixel

data on a logarith-

mic scale instead of

a linear scale. While

this is easy to imple-

ment, it requires

much higher SNRs,

which again places

more challenging

requirements on

receiver sensitivity

and illumination

source power.



contact losses that significantly degrade the Q-factor of on-chip passives (such as inductors,capacitors and transformers) and poorer devicegain and noise performance. Flicker noise isanother critical issue that can easily overwhelmthe low frequency detector and render the selec-tive noise filtering ineffective. CMOS detectorsalso demonstrated insufficient responsivity forvery fine temperature resolution (or noise equiv-alent temperature difference i.e. NETD). In [11]by using Schottky diodes in CMOS technology, aNEP of 40 pW/√Hz was demonstrated, whichwould take unreasonably long integration timeto resolve the high quality required by passiveimaging. In [12], a 90 GHz passive imager isimplemented in 65nm CMOS process, with aNEP of 0.11 pW/√Hz and a responsivity of 63kV/W.

Despite relatively poor silicon-based detectorperformance, simulation also indicates that aslong as the pre-amplifier has enough gain(>40dB), the detector noise becomes unimpor-tant and the system sensitivity (reflected byNETD) is only determined by the pre-amplifier’snoise contribution, as represented by the follow-ing equation [13]:

which specifies NETD being inversely propor-tional to front-end (LNA) bandwidth B, gain G,integration time T, and proportional to gainvariation ΔG and the system noise equivalenttemperature TSN. The gain variation ΔG is alsosignificant in CMOS technology due to supply,temperature variations and device flicker noise.To minimize ΔG and high flicker noise effect inCMOS, a Dicke switch was used to sample andcancel the gain variation. A Dicke switch is anSPDT switch placed in front of the imagingreceiver and alternates between sampling a ref-erence resistor and the incoming signal allowinglow frequency noises to be suppressed by cancel-lation. The detected power is also averaged overa certain integration time to further suppressreceiver noise. Dicke switches are typically oper-ated at a higher clock frequency than that of theNMOS flicker noise corner to further eliminateits effect on NEDT.

Figure 6a depicts the CMOS passive mm-wave imager in [14]. It includes both RF/analog

front-end and digital back-end. The RF/analogfront-end contains an input balun, a differentialLNA and detector, filter, PGA, AD convertertogether with an on-chip Dicke switch. The digi-tal back-end contains a digital multiplier, anintegrator and the required digital signal pro-cessing. This architecture places a multiplier andan integrator in the digital domain to suppresstheir flicker noise effect. All circuit related flick-er noise, including ADC’s, can be sampled andcancelled through the digital multiplier and inte-grator.

Several challenges reside in passive imagerdesign. First is the design of a high gain, lownoise and wide bandwidth LNA, which sets thestage for the overall image system sensitivity andresponse time. Reference [14] presented a designapproach to optimize the LNA performance witha demonstrated noise figure of less than 8 dB.The Dicke switch is another crucial componentwith its noise and loss directly affecting thereceiver noise dB by dB. In [14], we demonstrat-ed a Dicke switch structure based on a fully dif-ferential architecture (Fig. 6a) to minimize theinsertion loss and facilitate 100GHz imager oper-ation by placing the switch in the return path ofthe balun, instead of placing the switch directly inseries with the signal path. Figure 6c shows thedie photo of the integrated W-band passive mm-wave imager with an LNA, Dicke Switch, detec-tor and filter in 65 nm CMOS with a measuredNEP of 23fW/√

——Hz and 26fW/√

——Hz without or with

Dicke switch, respectively. The measured respon-sivity is larger than100 MV/W. When integratingover 30 ms time, the derived NETD is about 2 Kby multiplying with an additional factor of 2 toaccount for the Dicke switch effect of only half ofthe clock cycle for signal detection.

While CMOS mm-wave passive imagers havedemonstrated promising performance of RFfront-end [12–14], no CMOS passive imager sys-tem with fully integrated front and back endcapable of image capture been demonstrated todate. Assembly and packaging of mm-wavereceivers also requires further work. The currenton-chip or in-package antenna operates ineffi-ciently at mm-wave frequencies, which directlyimpacts the passive imager SNR margin. Aphased array can improve receiving signal SNRthrough beam forming techniques and has thepotential to relax the tight link budget. An addi-tional issue is how to suppress the on-chip cou-pling noise and interference from other parts ofthe system, due to supply and ground coupling,common mode noise and clock leakage, etc. Inthis regard, fully differential circuit implementa-tion and other isolation techniques become nec-essary. Many additional noise and interferencesources in the environment may also overwhelmthe small blackbody radiation from a distant tar-get and hinder the imaging resolution. Phasedarrays with sharp receiving beams can relax theissue with tightened power and area budget overeach imaging element.

SUMMARYWhile CMOS mm-wave receivers have becomemature for communication purposes, muchwork is still needed to meet the stringent

NETD TB

G

GSNT

=⎛

⎝⎜⎞

⎠⎟+ ⎛⎝⎜

⎞⎠⎟

1 2Δ

TTaabbllee 11.. Performance comparison of silicon mm-wave imaging receivers.

Feature [5] [7] [8] [9] (this work)

NEP (pW/ √Hz) 400 0.013 0.2 0.0013

Area (mm2) 0.03 1.2 0.21 0.0131

Power (mW) 5.5 200 93 13.5

Freq. (GHz) 600 94 94 183

Technology 0.25umCMOS 0.13um SiGe 65nm CMOS 65nm CMOS



demands of forming high resolution and lowpower mm-wave imaging arrays. While detectorbased imaging has matured to the point wherefull array active imagers are possible for short-range transmission modes, there is still no clearsolution for long-range reflective or backscatterimaging in CMOS technology. Receiver sensitiv-ity, noise performance and carrier frequencymust further improve while maintaining a strictpower and area budget to make full array reflec-tive systems feasible in CMOS. For passiveimaging, no fully integrated CMOS systemscapable of image capture have been demon-strated yet, which is mainly due to the stringentSNR margins and high susceptibility to interfer-ece associated with capturing and imaging ther-mal radiation. Even with these challenges, thelow power and high level of integration offeredby CMOS is anticipated to enable interestingnew applications including covert surveillance,low cost, and even portable mm-wave imagingsystems which are not achievable with III-Vimaging technology.

REFERENCES[1] E. R. Brown et al., “Fundamentals of Terrestrial Millime-

ter-wave and THz Remote Sensing,” Terahertz SensingTechnology Volume II: Emerging Scientific Applicationsand Novel Device Concepts, 2003, World Sci.

[2] K. B. Cooper et al., “Penetrating 3-D Imaging at 4 and25m Range Using a Submillimeter-Wave Radar,” IEEEMTT, vol. 56, no. 12, Dec 2008, pp 2771-2778

[3] J. J. Lynch et al., “Passive Millimeter Wave ImagingModule with Pre-Amplified Zero Bias Detection,” IEEETrans. Microw. Theory Tech., vol. 56, no. 7, July 2008,pp. 1592–600.

[4] H. Kazemi et al., “Ultra sensitiveErAs/InAlGaAs DirectDetectors for Millimeter Wave and THz Imaging Appli-cations,” IEEE/MTT-S Int’l. Microwave Symp., Honolulu,HI, Jun. 3-8, 2007, pp. 1367–70.

[5] E. Ojefors et al., “A 0.65 THz Flocal-Plane Array in aQuarter-Micron CMOS Process Technology,” IEEE JSSC,Vol. 44, No. 7, July 2009, pp. 1968–76.

[6] E. Ojefors and U. R. Pfeiffer, “ A 650GHz SiGe ReceiverFront-End for Terahertz Imaging Arrays,” IEEE ISSCC,Feb. 2010, pp. 430–31.

[7] L. Gilreath et al., “A 94-GHz Passive Imaging ReceiverUsing A Balanced LNA with Embedded Dicke Switch,”IEEE RFIC, May 2010, pp. 79–82.

[8] K. W. Tang et al., “65-nm CMOS, W-Band Receivers forImaging Applications,” IEEE CICC, Sept. 2007, pp749–52.

[9] A. Tang and M.-C. F. Chang, “183GHz 13.5mw/pixelCMOS Regenerative Receiver for mm-Wave ImagingApplications,” IEEE ISSCC, vol. 54, Feb. 2011, pp.296–97.

[10] A. Tang et al., “A 144 GHz 2.5mW Multi-Stage Regen-erative Receiver for mm-Wave Imaging in 65nm CMOS,”IEEE RFIC 2011.

[11] R. Han et al., “280-GHz Schottky Diode Detector in130-nm Digital CMOS,” IEEE CICC, Sept. 2010

[12] A. Tomkins, P. Garcia, and S. P. Voinigescu, “A PassiveW-band Imaging Receiver in 65-nm Bulk CMOS,” IEEE J.Solid-State Circuits, vol. 45, no.10, Oct. 2010.

[13] J. W. May and G. M. Rebeiz, “Design and Characteri-

FFiigguurree 66.. a) One CMOS passive mm-wave imager structure; b) schematic of full differential cascadedLNA; c) die photo of fabricated W-band CMOS passive imager.

Out

put

to d

etec

tor

Intp

ut f

rom

Bal

un

Stage 4

Out

put

Intp

ut A/D[ ]2 nn

Digital

Digitalintegrator

Digitalmultiplier

PGAFilter

RF and analog

LNA

Switch clock

Balun Detector

(a)

(b)

(c)

1070um45

7um

1ƒτ

Σ

Stage 3Stage 2Stage 1

DickeSW

Even with these chal-

lenges, the low

power and high level

of integration

offered by CMOS is

anticipated to enable

interesting new

applications includ-

ing covert surveil-

lance, low cost, and

even portable mm-

wave imaging sys-

tems which are not

achievable with III-V

imaging technology.



zation of W-Band SiGe RFICs for Passive Millimeter-Wave Imaging,” IEEE Trans. Microw. Theory Tech., vol.58, no. 5, May 2010.

[14] Q. J. Gu et al., “A 100 GHz Integrated CMOS PassiveImager with >100MV/W Responsivity, NEP,” IET Elec-tronics Letter, vol. 47, issue 9, 2011, Featured Paper,pp. 544–45.

BIOGRAPHIESADRIAN TANG is currently a Ph.D. student at the University ofCalifornia, Los Angeles (UCLA) with research focusing onactive sub-mm-wave imaging and mm-wave imaging radarsystem design.

QUN JANE GU [S’00, M’07] received her B.S. and M.S. fromHuazhong University of Science and Technology, Wuhan,China, in 1997 and 2000, an M.S. from the University ofIowa, Iowa City, in 2002 and her Ph.D. from UCLA in 2007,all in electrical engineering. She received a UCLA fellowshipin 2003 and a Dissertation Year Fellowship in 2007. Aftergraduation, she worked as a senior design engineer in theWionics Realtek research group and a staff design engineerin AMCC on CMOS mm-wave and optic I/O circuits. Mostrecently, she was a postdoctoral researcher in UCLA. InAugust 2010 she joined the University of Florida as anassistant professor. Her research interest spans high-effi-ciency low-power interconnect, mm-wave and sub-mm-wave integrated circuits and SoC design techniques, as wellas integrated terahertz imaging systems.

FRANK CHANG is the Wintek Endowed Chair and Distin-guished Professor of Electrical Engineering and Chairmanof the Electrical Engineering Department, UCLA. Beforejoining UCLA, he was the assistant director and depart-ment manager of the High Speed Electronics Laboratory atRockwell Science Center (1983–1997), Thousand Oaks, Cali-fornia. In this tenure, he developed and transferred theAlGaAs/GaAs heterojunction bipolar transistor (HBT) andBiFET (planar HBT/MESFET) IC technologies from theresearch laboratory to the production line (now ConexantSystems and Skyworks). HBT/BiFET productions have growninto multi-billion-dollar businesses and dominated the cellphone power amplifiers and front-end module markets(currently exceeding one billion units/year). Throughout hiscareer, his research has primarily focused on the develop-

ment of high-speed semiconductor devices and integratedcircuits for RF and mixed-signal communication and imag-ing system applications. He was the principal investigatorat Rockwell in leading DARPA’s ultra-high speed ADC/DACdevelopment for direct conversion transceiver (DCT) anddigital radar receivers (DRR) systems. He was the inventorof the multiband, reconfigurable RF-interconnects, basedon FDMA and CDMA multiple access algorithms, for Chip-Multi-Processor (CMP) inter-core communications andinter-chip CPU-to-Memory communications. He also pio-neered the development of world’s first multi-gigabit/secADC, DAC and DDS in both GaAs HBT and Si CMOS tech-nologies; the first 60GHz radio transceiver front-end basedon transformer-folded-cascode (Origami) high-linearity cir-cuit topology; and the low phase noise CMOS VCO(F.O.M.<-200dBc/Hz) with Digitally Controlled on-chip Arti-ficial Dielectric (DiCAD). He was also the first to demon-strate CMOS oscillators in the terahertz frequency spectrum(1.3 THz) and the first to demonstrate a CMOS activeimager at the sub-mm-Wave spectra (180GHz) based on aTime-Encoded Digital Regenerative Receiver. He was alsothe founder of an RF design company G-Plus (now SST andMicrochip) to commercialize WiFi 11b/g/a/n power ampli-fiers, front-end modules and CMOS transceivers. He waselected to the US National Academy of Engineering in2008 for the development and commercialization of GaAspower amplifiers and integrated circuits. He was also elect-ed as a Fellow of IEEE in 1996 and received IEEE DavidSarnoff Award in 2006 for developing and commercializingHBT power amplifiers for modern wireless communicationsystems. He was the recipient of 2008 Pan Wen YuanFoundation Award and 2009 CESASC Career AchievementAward for his fundamental contributions in developingAlGaAs/GaAs hetero-junction bipolar transistors. His recentpaper “CMP Network-on-Chip Overlaid with Multiband RF-Interconnect” was selected for the Best Paper Award in2008 IEEE International Symposium on High-PerformanceComputer Architecture (HPCA). He received Rockwell’sLeonardo Da Vinci Award (Engineer of the Year) in 1992;National Chiao Tung University’s Distinguished AlumnusAward in 1997; and National Tsing Hua University’s Distin-guished Engineering Alumnus Award in 2002. He earnedhis B.S. in physics from National Taiwan University in 1972,his M.S. in materials science from National Tsing Hua Uni-versity in 1974, and his Ph.D. in electronics engineeringfrom National Chiao Tung University in 1979.


cmos receivers for active and passive mm-wave...

Documents