A picosecond time-resolved photoluminescence microscope with detection at wavelengths greater than 1500 nm

Transcription

1 REVIEW OF SCIENTIFIC INSTRUMENTS VOLUME 72, NUMBER 5 MAY 2001 A pcosecond tme-resolved photolumnescence mcroscope wth detecton at wavelengths greater than 1500 nm J. M. Smth, a) P. A. Hskett, I. Gontjo, b) L. Purves, and G. S. Buller Department of Physcs, Herot-Watt Unversty, Rccarton, Ednburgh EH14 4AS, Unted Kngdom Receved 1 December 2000; accepted for publcaton 19 February 2001 We descrbe a pcosecond resoluton tme-resolved photolumnescence mcroscope wth hgh detecton senstvty at wavelengths extendng beyond 1500 nm. The nstrument performs tme-correlated sngle photon countng usng an InGaAs/InP sngle photon avalanche dode as a detector, and provdes temporal resoluton of less than 300 ps full wdth at half maxmum and spatal resoluton down to 4 m at a sample temperature between 4 and 300 K. Analyss of nose characterstcs ndcates the ablty to measure the excess carrer lfetmes of semconductor devces wth excted carrer denstes of less than cm Amercan Insttute of Physcs. DOI: / I. INTRODUCTION Tme-resolved photolumnescence TRPL measurements allow the nvestgaton of electronc relaxaton mechansms n optcally actve materals. In semconductors, for nstance, the lumnescence ntensty depends upon the excess carrer densty, whch reduces on a tme scale rangng from pcoseconds to mcroseconds, as a result of ntraband carrer dynamcs and varous nterband recombnaton processes. Understandng of the domnant processes n the relaxaton s of great value n the characterzaton of materal qualty and desgn of new devces. Tme-correlated sngle photon countng TCSPC 1 usng sngle photon avalanche dodes SPADs offers TRPL measurement of unrvaled senstvty and wth pcosecond tme resoluton, but has untl now been lmted to detecton wavelengths shorter than 1500 nm, ths beng the long wavelength lmt of senstvty for Ge homojuncton devces. 2,3 The lowloss telecommuncatons wndow between wavelengths of 1500 and 1600 nm, n whch most long haul telecommuncatons devces are desgned to operate, has therefore not been accessble, except by less senstve methods such as pump-probe 4 or up-converson detecton. 5 As a result of extensve nvestgatons nto the use of commercal InGaAs/InP separate absorpton, gradng, and multplcaton layer SAGM avalanche photododes as photon counters, 6 we have now dentfed devces whch can be operated n photon countng mode to offer nose equvalent powers NEPs as low as WHz 1/2 at an operatng temperature of 77 K, combned wth tmng resoluton of less than 300 ps. We have ncorporated one such detector nto a TRPL mcroscope, and performed prelmnary measurements on a semconductor devce structure that ndcates senstvty to excess carrer denstes as low as cm 3. 7 Ths artcle descrbes n detal the mcroscope hardware, the measurement method, and key performance ssues. It a Electronc mal: j.m.smth@hw.ac.uk b Current address: Department of Electrcal Engneerng, UCLA, Los Angeles, Calforna provdes a more detaled, general analyss of the senstvty of TCSPC to low lght levels, and derves some results that are both helpful n optmzng ths nstrument, and whch ndcate that senstvty to sgnal ntenstes even lower than those demonstrated prevously 7 are measurable. II. APPARATUS AND OPERATION A schematc dagram of the TRPL apparatus s shown n Fg. 1. For ease of vewng, the electrcal confguraton prmarly occupes the left-hand sde of the dagram and the optcal confguraton the rght. The pulsed optcal source s a passvely Q-swtched InGaAsP/InP semconductor dode laser emttng at a wavelength of 1305 nm. The Q-swtchng mechansm s based on an effcent saturable absorber formed by heavy on bombardment, whch permts relable producton of optcal pulses of duraton 15 ps and pulse energy of 5 pj. 8 Such lasers have been produced for operaton at numerous other wavelengths, ncludng 800 and 1450 nm. The exctaton laser dode s drven by an Avtech AVO-9 electrcal pulse generator, whch also provdes a synchronzed trgger for a second pulse generator unt, a Hewlett Packard HP81110A, to supply both the start pulse to the tme-to-ampltude converter TAC va a constant fracton dscrmnator CFD and a 5 V gate pulse to bas the cooled InGaAs/InP SPAD detector nto the Geger mode. The SPAD s otherwse held at a dc bas just below ts breakdown voltage approx 30 V at 77 K. A varable tme delay n the Avtech pulse generator s adjusted to ensure that the optcal sgnal arrves at the detector durng the gated-on perod. The optcs for ths mcroscope are mounted on a purpose-bult baseplate, whch can be postonally adjusted, usng mcrometer precson actuators, relatve to the sample, whch s fxed. Ths slotted baseplate technology was developed for prevous experments n optcal nterconnects 9 and supports the stable postonng of optcal components mounted n cylndrcal modules. Its ablty for rapd optcal reconfguraton makes ths nstrument deal for the study of a /2001/72(5)/2325/5/$ Amercan Insttute of Physcs Downloaded 10 Jan 2002 to Redstrbuton subject to AIP lcense or copyrght, see

2 2326 Rev. Sc. Instrum., Vol. 72, No. 5, May 2001 Smth et al. FIG. 2. Instrumental response ( PL 1305 nm) and sample PL decay from an InGaAsP semconductor heterostructure ( PL 1525 nm) usng the TRPL mcroscope. FIG. 1. Schematc dagram of the TRPL mcroscope. For ease of vewng, the electrcal and optcal confguratons occupy the left- and rght-hand sdes of the dagram, respectvely. wde range of samples, whereby frequent nterchange between dfferent exctaton sources, passve optcal components, and detectors may be requred. Polarzng beamspltters coated for nm wavelength operaton and a 90 polarzaton rotator combne to drect the collmated exctaton pulse towards the sample, and t s focused onto the sample surface usng one of a selecton of mcroscope objectves. The same lens s used to collect the now unpolarzed lumnescence, whch s then fltered spectrally usng nterference flters 50% peak transmsson, 12 nm full wdth at half maxmum FWHM and attenuated as necessary, and focused nto a sngle mode optcal fber for delvery to the detector. The choce of objectve lens used at the sample depends upon the desred spatal resoluton and collecton effcency, and upon whether the sample s to be placed n the contnuous flow, vacuum loadng cryostat Oxford Instruments CF1104 n whch case the mnmum workng dstance s lmted to about 5 mm. Usng a 36 reflectng objectve lens wth a numercal aperture NA of 0.5 we can currently obtan 4 m spatal resoluton under these condtons. The fber nto whch the lumnescence s focused couples to the sngle mode pgtal of the SPAD, whch feeds nto the top of an Oxford Instruments Optstat DN exchange gas cryostat contanng the detector. Ths allows control of the detector temperature T SPAD, a parameter that s crtcal n determnng ts nose characterstcs and maxmum operaton frequency, down to 77 K. Pror to the lumnescence measurement, algnment of the nstrument s smplfed by shnng lght through the collecton fber n the reverse drecton and vewng both the resultant mage of the detecton aperture and the exctaton spot on the sample surface usng an nfrared camera. Ths step has the addtonal advantage of revealng drectly the relatve szes of the exctaton and detecton areas, whch can be mportant n the avodance of carrer dffuson effects n the tme-resolved measurements. As well as beng gated by the HP pulse generator, the SPAD s passvely quenched usng a 100 k resstor also mounted nsde the cryostat to mnmze the flow of current through the detector upon avalanche breakdown. Ths serves to lmt the fllng of deep traps n the semconductor materal, from whch subsequent carrer release s thought to be responsble for trggerng unwanted afterpulses durng the next detector gate. 10 The voltage across a further 50 resstor n seres wth the SPAD and quench resstor s amplfed and used as the stop sgnal of the TAC va a second CFD. The ampltude of the voltage spkes that occur across the 50 resstor due to the fast transents n the detector gatng s reduced below that of the avalanche events by settng the gate rse tme to 20 ns. It s the ablty to adjust the gate transent tme that earns the HP81110A ts place n ths apparatus. Havng establshed the mnmum rse tme for successful operaton, t should be possble to replace the HP81110A wth a more basc pulse generator combned wth a smple low-pass flter to acheve the same result. The CFDs, TAC, and multchannel analyzer MCA used to generate the data hstogram are convenently ntegrated on an Ednburgh Instruments TCC900 data acquston card, runnng on Wndows n a PC. III. TEMPORAL RESOLUTION The nstrumental response of the complete mcroscope, wth a detector temperature of 77 K and exctaton repetton rate of 1 khz, s shown n Fg. 2. Ths hstogram was acqured by placng a mrror n place of the sample and recordng photons from the reflected and sutably attenuated exctaton pulse, and so the wavelength of lght beng detected s that of the exctaton laser, 1305 nm. Also shown n the fgure s a porton of an example TRPL hstogram measured from an InGaAsP multple quantum well heterostructure at a detecton wavelength of 1525 nm. The decay takes the form of a sngle exponental supermposed wth both Possonan and temporally correlated nose, the orgns of whch wll be dscussed n the next secton. Downloaded 10 Jan 2002 to Redstrbuton subject to AIP lcense or copyrght, see

3 Rev. Sc. Instrum., Vol. 72, No. 5, May 2001 Photolumnescence mcroscope 2327 The key feature of the nstrumental response s the full wdth at half maxmum FWHM of 230 ps. Ths s determned prncpally by the tmng jtter of the SPAD, snce the combned jtter of the other tmng components amounts to at most a few tens of pcoseconds. The physcal orgn of the SPAD jtter s a matter of debate, 11,12 but t s a weak functon of the wavelength of lght to be detected as a result of the absorpton spectrum of the InGaAs layer, wth 1550 nm lght ncreasng the jtter to 270 ps. Iteratve reconvoluton of the measured lumnescence data from the nstrumental response can n practce facltate the fttng of decay lfetmes shorter than the FWHM. Whle the maxmum extent to whch ths can be acheved s subject to queston, 1 a number of authors have demonstrated satsfactory fttng to data wth decay lfetmes of around one ffth of the nstrumental FWHM, 13,14 and on ths bass we estmate the ultmate tmng resoluton of ths mcroscope, under deal condtons, to be 60 ps. Ths resoluton s, however, an upper lmt when the nfluence of nose on the sgnal s neglgble. Ths s rarely the case, and ndeed s often undesrable snce such a stuaton generally only occurs at the expense of long measurement tmes. For ths reason, t s better to approach each measurement ndvdually: by knowng what accuracy s requred, and wth some a pror knowledge of the photolumnescence decay characterstcs, we can determne, and to some extent mnmze, the length of tme that each measurement wll take to perform. To ths end, the followng secton contans a descrpton of the orgns and manfestatons of nose n these measurements, and of how ths understandng can be used to optmze nstrumental performance. Note that much of what follows s applcable to any TRPL measurement usng TCSPC, but that t s of partcular mportance here due to the lmtatons placed on the operatng frequency of ths detector. IV. SENSITIVITY AND NOISE ANALYSIS The TCSPC hstogram s the aggregate of all the temporally correlated sgnal counts and the uncorrelated background counts regstered durng the acquston tme t acq. Temporally uncorrelated background counts orgnate mostly from two sources; stray photons ncdent on the detector and dark counts, whch are avalanche events caused not by photogenerated carrers, but rather by carrers that have been generated thermally or by trap release n the detector. In the hypothetcal case of a noseless hstogram, the background counts are equal n number n each channel, and so can be smply subtracted from the acqured data to leave only the sgnal. Such a smple treatment s n practce hndered by two factors: temporally correlated nose, manfested as nhomogenety n the detecton probablty across the gate and Possonan nose. The latter s a result of the stochastc nature of TCSPC and s equal n magntude to the square root of the number of counts n a gven channel. The former can result from a varety of effects; a nonunform gate voltage, electromagnetc pckup n the detecton crcut, and nonlnearty n the data acquston hardware beng the most lkely causes. The detecton probablty functon, or gatng FIG. 3. A typcal gatng functon data set wth a count level of 10 4 and thus a relatve Possonan nose ampltude of 1%. A gate duraton of 200 ns has been used wth a gate transent tme of 20 ns. The earler part of the gate s seen to be less unform, due to post-transent rngng n the detector crcut. functon, can be measured drectly by llumnatng the gated SPAD wth temporally uncorrelated photons and acqurng a large enough data set that Possonan nose, as a fracton of the total number of counts, may be neglected. An example of such a data set, for a gate of wdth 200 ns, s shown n Fg. 3. Each subsequent data set can be normalzed by dvdng through by the gatng functon. To follow ths process relgously s somewhat arduous, however, snce for best results the gatng functon must be measured drectly pror to each set of measurements under a gven set of condtons detector temperature, gate wdth, MCA channel wdth, etc., and can take some tme to acqure. It s clearly preferable to take care n generatng a flat, unform gatng functon when constructng the apparatus, so that all but the most demandng of measurements can be performed wthout the need for ths adjustment. Even when the above precautons are taken, the gate normalzaton procedure does not n practce account for all of the temporally correlated nose, due to slowly varyng elements such as drft of the exctaton laser and n the data acquston chan. What remans s a reduced temporally correlated nose component, wth an ampltude that can be represented as a constant fracton of the count level n a gven channel. In our prelmnary nvestgatons usng the apparatus descrbed, s approxmately 0.1. It s hoped that ths value can be reduced sgnfcantly n our apparatus wth the adopton of an optcal trgger to start the TAC, whch wll elmnate from the tme measurement crcut the drft between the optcal pulse and the electrcal trgger wthn the laser pulse generator. Obtanng a small value of s mportant, snce, as we shall see n the followng analyss, t alone determnes the ultmate senstvty of our measurements. We start our nose analyss by defnng the sgnal-tonose rato SNR n the th channel of the MCA as the rato of the number of sgnal counts to the sum of the Possonan nose and the temporally correlated nose n that channel: Downloaded 10 Jan 2002 to Redstrbuton subject to AIP lcense or copyrght, see

4 2328 Rev. Sc. Instrum., Vol. 72, No. 5, May 2001 Smth et al. SNR N sg N sg r bkgnd tn rep N sg r bkgnd tn rep, where r bkgnd s the normalzed background count rate, t s the tme wdth per MCA channel ndeed r bkgnd / t s, by defnton of r bkgnd, the probablty of a background count occurng n a sngle tme wndow of duraton t, and n rep s the number of pulse repettons performed n the measurement. 15 The number of sgnal counts n the th channel of the MCA can be expressed as a fracton,, of the peak value. For a sngle exponental decay wth a lfetme that s slow compared wth the nstrumental response tme, the resultng expresson s N sg t n rep P phot, where P phot s the probablty of measurng a sgnal photon n any gven pulse. For a decay that s faster than the nstrumental response FWHM, and therefore requres reconvoluton analyss, a good approxmaton s acheved by replacng t wth the FWHM n Eq. 2. Equaton 1 llustrates the fact that as data s acqured, Possonan nose reduces relatve to the temporally correlated components, and n the lmt of large n rep the temporally correlated nose s domnant. Snce no more than a factor of 2 n the SNR s to be ganed by contnung to accumulate data beyond the pont at whch the two nose components are equal n magntude, we choose ths condton to provde a practcal ndcaton of the achevable senstvty, and the number of pulsed exctatons requred to attan t. Usng Eqs. 1 and 2, the SNR s thus gven by P phot SNR nst 2 P phot r bkgnd and the correspondng number of pulsed exctatons s n rep 2 t P phot r bkgnd. 4 Note that the SNR accordng to Eq. 3 cannot exceed 1/2. The nherent dffculty n fttng decay lfetmes to hgh accuracy as a result of ths lmt must be compensated for by fttng over a broad range of. More quanttatvely, the attanable accuracy of the ft to data spannng a gven range of s approxmately mn /(SNR mn max ), where mn and max defne the range of and SNR mn s the SNR at mn. For example, a SNR of 1 at 0.1 would enable a decay tme constant to be ftted to the peak of a sgnal wth an accuracy of 10%. It should be noted that, n the case of a decay that s fast compared wth the nstrumental response and therefore requres deconvoluton analyss, sgnfcantly hgher SNR, or fttng over a correspondngly greater range of would be requred to acheve a smlar accuracy. Wth the above result n mnd, Eq. 3 permts an estmate of the mnmum sgnal photon detecton probablty for whch such accuracy can be obtaned. For an example stuaton of a monomolecular decay wth 10 ns, and usng r bkgnd 100 s 1, the resultng photon detecton probablty s FIG. 4. Plot of r bkgnd aganst and f rep for T SPAD 77, 120, and 160 K. Closed shapes are wth the detector unlt and open shapes are wth the detector llumnated to gve a total TAC stop probablty of 5%. P phot Ths value compares favorably wth the value of P phot 10 5 deduced from the smpler model n Ref. 7. In an undoped semconductor sample of planar geometry, exhbtng monomolecular photolumnescence decay, we estmate that such a detecton probablty would result from an excted carrer densty of about cm Equaton 4 can be used to estmate the length of tme taken to perform a gven measurement, t acq, by dvdng through by the pulse repetton frequency f rep. However, the background count rate r bkgnd s strongly dependent on f rep due to the afterpulsng effect descrbed earler. Fgure 4 shows ths dependence for 77 K T SPAD 160 K under two dfferent llumnaton condtons. The frst of these, shown n closed shapes, was measured wth the detector n the dark to represent the weak sgnal lmt; the second, shown n open shapes, was measured wth a total detecton probablty per gate pulse of 0.05, whch we have chosen to represent the allowable strong sgnal lmt, pursuant to the avodance of pulse pleup. 1 In the weak sgnal lmt at a detector temperature of 77 K, Fg. 4 shows that afterpulsng starts to become mportant at a pulse repetton frequency of around 5 khz. At frequences or temperatures hgher than ths, the ultmate senstvty of the nstrument wll be reduced below that dscussed above, correspondent wth the ncreased r bkgnd due to afterpulsng or thermally generated counts, respectvely. Note that at a fxed f rep, afterpulsng s reduced as T SPAD s ncreased, a characterstc attrbutable to the thermonc nature of trap emsson. The competton between ths effect and the ncreased rate of thermally generated counts results n an optmum temperature that mnmzes r bkgnd for a gven f rep. Estmatng t 1 ns as the longest sutable MCA bn tme wdth for measurng a 10 ns decay to 0.1 wth 10% accuracy, the shortest tme n whch the above measurement can be performed s estmated to be s, or 44 h. Whle ths may be regarded as an nordnately long tme for a measurement, t s encouragng to note that by ncreasng P phot by a factor of 4 to 10 5 correspondng to a semconductor excess carrer densty of cm 3, Eqs. 1 and 2 Downloaded 10 Jan 2002 to Redstrbuton subject to AIP lcense or copyrght, see

5 Rev. Sc. Instrum., Vol. 72, No. 5, May 2001 Photolumnescence mcroscope 2329 predct that the same 10% accuracy n lfetme fttng can be acheved wth t acq reduced to 1.7 h. For sgnal ntenstes larger than that correspondng to P phot 10 5, sgnfcant gans n measurement speed can be acheved by ncreasng the detector temperature. As a rule of thumb, t acq s mnmzed by settng T SPAD to maxmze the f rep that gves a background count level equal to the sgnal count level: r bkgnd P phot. 5 For example, f P phot 10 3 and all other parameters are unchanged, the shortest acquston tme corresponds approxmately to r bkgnd 10 khz, whereby the fastest repetton frequency accordng to Fg. 4 s approxmately 40 khz at T SPAD 120 K gvng t acq 8s. Note that afterpulsng characterstcs may vary greatly from detector to detector, and so Fg. 4 should be generated afresh for a gven SPAD. The SPAD detecton effcency SPAD s about 8% at T SPAD 77 K and PL 1550 nm. A complete calbraton of the detecton effcency spectrum of these detectors has not yet been performed, but they are expected to behave smlarly n ths respect to other InGaAs/InP SAGM devces, n whch SPAD falls off rapdly at optcal wavelengths longer than the In 0.53 Ga 0.47 As absorpton edge. 10 As regards the short wavelength lmt, a substantal ncrease n tmng jtter to 1 ns observed for PL 1000 nm s attrbuted to the onset of absorpton n the n-type InP contact layer, characterstc of backllumnaton of a SAGM devce. ACKNOWLEDGMENTS Ths work s funded by the UK Engneerng and Physcal Scences Research Councl ref GR/L81895 and by the European Commsson under the EQUIS project IST The authors would also lke to thank Dawn Marshall and Alan Mller of St. Andrews Unversty for supplyng test samples and Paul Townsend of Cornng Research Center Ipswch, UK for nvaluable dscussons. 1 D. V. O Connor and D. Phllps, Tme-Correlated Sngle Photon Countng Academc, London, A. Lacata, S. Cova, F. Zappa, and P. A. Francese, Opt. Lett. 18, G. S. Buller, S. J. Fancey, J. S. Massa, A. C. Walker, S. Cova, and A. Lacata, Appl. Opt. 35, R. Takahash, Y. Kawamura, T. Kagawa, and H. Iwamura, Appl. Phys. Lett. 65, A. D. Güçlü, C. Rejeb, R. Macejko, D. Morrs, and A. Champagne, J. Appl. Phys. 86, P. A. Hskett, G. S. Buller, J. M. Smth, A. Y. Loudon, I. Gontjo, A. C. Walker, P. D. Townsend, and M. J. Robertson, Appl. Opt. 39, J. M. Smth, P. A. Hskett, and G. S. Buller, Opt. Lett. accepted. 8 Z. I. Alferov, A. B. Zuravlev, E. L. Portno, and N. M. Stel makh, Sov. Tech. Phys. Lett. 12, F. B. McCormck, F. A. P. Tooley, T. J. Cloonan, J. L. Brubaker, A. L. Lentne, R. L. Morrson, S. J. Hnterlong, M. J. Herron, S. L. Walker, and J. M. Sasan, Appl. Opt. 31, A. Lacata, F. Zappa, S. Cova, and P. Lovat, Appl. Opt. 35, A. L. Lacata, S. Cova, A. Spnell, and F. Zappa, Appl. Phys. Lett. 62, A. Spnell and A. L. Lacata, IEEE Trans. Electron Devces 44, T. A. Lous, G. Rpamont, and A. Lacata, Rev. Sc. Instrum. 61, J. S. Massa, G. S. Buller, A. C. Walker, J. Smpson, K. A. Pror, and B. C. Cavenett, Appl. Phys. Lett. 64, A smlar treatment of the SNR has been used for analyss of TCSPC n a tme-of-flght rangng system; see S. Pellegrn et al., Meas. Sc. Technol. 11, The parameters used n ths estmate are radatve recombnaton coeffcent cm 3 s 1 ; nonradatve recombnaton lfetme 10 ns; sample volume detected cm 3 ; refractve ndex of semconductor 3.5; objectve lens NA 0.4; and SPAD detecton effcency 8%. Downloaded 10 Jan 2002 to Redstrbuton subject to AIP lcense or copyrght, see