I E E E TR-ANS.ACTIONS ON ELECTROS DEVICES, VOL. ~ ~ 1 - 1 so. 6 , 1, J.ASLARY 1969 Large-Signal Analysis of a Silicon Read Diode Oscillator Abstract-Thispaperpresentstheoreticalcalculations of the large-signal admittance and efficiency achievable in a silicon p-n-v-n Read IMPATT diode. A simplified theory is employed to obtain a startingdesign.Thisdesignisthenmodifiedtoachievehigher efficiencyoperation as specificdevicelimitations arereachedin large-signal(computer)operation.Self-consistentnumericalsolutions are obtained for equations describing carrier transport, carrier generation,andspace-chargebalance.Thesolutionsdescribethe evolution in time of the diode and its associated resonant circuit. Detailed solutions are presented of the hole and electron concentrations,electric field, andterminalcurrentandvoltageatvarious points in time during a cycleof oscillation. Large-signal values of the diode’snegativeconductance,susceptance,averagevoltage,and power-generating efficiency are presented as a function of oscillation amplitudefora fixed averagecurrentdensity.Forthestructure studied,thelargest microwavepower-generatingefficiency (18 percent at 9.6 GHz) has been obtained at a current density of 200 A/cmZ, but efficiencies near10 percent were obtained over a rangeof current density from 100 to 1000 A/cm2. s I. INTRODUCTIOS The aim of thedetailedcalculation,inadditionto the enhancement of our understandingof the oscillation mechanisms, was to find conditions under which high efficiency and power outputcan be obtained,andto A establish theoretical limits for these quantities. rigorous optimization, however, was not attempted in thisexploratory n-ork. This would have had to start with carefully formulated constraints and would have constituted a very large-scaleproject.However,the largest efficiency obtainedinthedetailednumerical calculation reported here is 18 percent, in good agreement with the simplified theoretical estimate. The large-signal operating characteristics of the diode are calculated in the following two ways: A . VoltageDriven To study the performance at a given frequency, but for various bias and load conditions, it is convenient to drive the diode with a sinusoidal voltage applied through a coupling capacitor [see Fig. 1 (a)]. From such studies, the admittance at the fundamental frequency is obtained as a function of driving voltage amplitude. Likewise, the ac power delivered by the diode into the voltage generator is obtained. E L F - C O N S I S T E N T large-signalnumericalsolutionsareobtained for equations whichdescribe carrier generation and space-charge balance in a silicon p-n-v-n Readdiode [ l ] microwaveoscillator. The solutions describe the evolution in time of the diode and associated circuit (see Fig. 1). The doping profile of the diode (see Fig. 2) was evolved starting Ivith design considerations discussed in [2], and reviewed here, B . Free-RunningOscillator and modified afterinitialresults. for improvement of For another series of calculations,thediode is imefficiency. From the numerical solutions, the large-sigbeddedinaresonantcircuit [see Fig. l(b)] and pernal operating characteristics of the diode and the effi- forms as a free-running oscillator. In as much as the ac ciency atthefundamentalfrequency (defined as ac voltage in the oscillator case is nearly sinusoidal, both powerdelivered bythediodedividedbydc power calculationsshould,andindeeddo,givecomparable dissipated) are computed. results. Read in his original paper [l ] suggested that an effiIn Section 111,n.e present the large-signal operating ciency of 30 percent should be obtainable in a silicon characteristicsanddetailedsolutions.Theoperating diode.Thiscalculation,however,neglectedthefinite characteristics, i.e., admittance parameters, average width of the avalanche region, and neglected the differvoltage, and power generating efficiency, are shown as ence in hole and electron ionization coefficients in sili- a function of diodeacvoltageamplitude in Figs.4 con. A modification of Read’scalculation, which in- through 9. Fig. 10 is a diode admittance plot (small and cludes the finite width of the avalanche region and the large signal) which sho\vs the variation of conductance experimental values of ionization coefficients for silicon andsusceptancewithoscillationamplitudeandfre[ 3 ] , is presentedinSection 11. I t isfound thatthe quency. Also indicatedarethe power generating effiavalancheregionmusthaveanappreciablewidthfor ciencies. Fig. 10 is a compact presentation of the importhe diode tohave high power capabilityandhigh tant results of thispaper.Detailedsolutionsarepreefficiency. A theoretical efficiency of theorder of 15 sented of the hole and electron concentrations, electric percent is obtained for conditionswhichshould ap- field, and terminal current and voltage at various points proach maximum power output. in time during a cycle of oscillation. These solutions are sholvnin Fig. 11 ( J d , =200.4/cm2,frequency = 11.4 Manuscript received June 3, 1968. GHz) and Fig. 12 ( I d c = 1000 -4,’cm2, frequency The authors are with Bell Telephone Laboratories, Inc., hlurray Hill, N. J . = 13.4GHz).The d3,namics ofcharge-pulsebuildup Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. SCHARFETTER . W D GUMXIEL: SILICOK RE;\D DIODE 0SCILL.ATOR 65 one-half the average voltage l,,D developed across the drift region [ l ] , [ 2 ] . L\tthedriftfrequency (f~=onehalf theinversetransittime of thedriftregion),the motion of Qmaxresults under favorable conditions in an ac particle current \vhich is = 180 degrees out of phase \vith the ac voltage across the diode. The average of the particle current is the dc current Jdc. The particle current slving is, therefore, a t most from zero to twice the dc current. For a square wave of particle current and a sinusoidal variation of drift voltage, both with magnitudeandphaseasdescribedabove,themicro\vave poi\-er-generating efficiency is +TrTIG (a) (b) \\-here the total dc voltage is divided betxveen the part developedacrosstheavalanche region I'dcA andthe part developed across the drift region T-~,D. S o t e t h a t \ve are neglecting any appreciallle average ac po\ver contribution from the avalanche region; this is because the IO" avalanche region voltage is inductively reactive relative to the particle current [4]. The displacement current is capacitively reactive relative to the diode voltage and therefore contributes no average ac pon-er. T h e use of a simple equivalent circuit [4] for the avalanche region, \vhich is derived from small-signal considerations, may of course have no validity under large-signal conditions. However, n-e will use this approach as a starting point for a simple 'design theor>-, and compare detailed numerical calculations \\-ith the simple treatment. I n Read'scalculation, it \\-as assumedthat no dc po\ver loss \\-as associated \\it11 theavalancheregion N+ N P since it n-as considered small in extent cornpared to the drift region, and that such an avalanche regioncould 1 0 ~ ~ I~ I' 11 1 I 1 1 1 2 3 4 5 6 7 8 9 1 0 produce an ac particle current amplitude equal to the DISTANCE IN MICRONS dc current \\-henoperated at the drift frequency and Fig. 2 . Netimpurityconcentration of Readdiode 7A versus distance. operated I v i t h ac voltage amplitude equal to one-half thedcvoltage[sameassumptionsas used above in deriving ( I ) ] . Since TTdcA a-as neglectedcompared to i n theavalanche region anddriftthroughthedrift 1 7 d c D , Read predictedan efficiency of 1,'n or abut region is in qualitative agreement \\it11 Read's original 30 percent. prediction of operation. Small-signalcalculations [ I ] , [ 2 ] , [4], [SI shou. t h a t Section 11: presentsconclusions,andtheAppendix aresonancefrequerqf x is associated \\it11 an?' avdescribes the time evolution computer program. alanche region, and that it increases as the square root of dccurrent. To obtain small-signalnegative resis11. SIMPLIFIED DIODE DESIGN tancerequiresoperationahovethisfrequency. Ho\vIn this section, we extend Read's efficiency calcula- ever, operation of an avalanche region a t a frequency tion toinclude thedcvoltagedevelopedacrossthe \vel1 above its resonance frequenq. \vas found to result avalanche region, and to allow for the unequal ionizainvery inefficient oscillationsinalimitednumber of tion rates of holes andelectrons, as appropriate for exploratorylarge-signalcalculations t h a t \\-erecomsilicon [3]. As pointed out by Read, high-power operapleted on a variety of Read diode structures. Small-sigtion requires the generation of as large a charge pulse nal calculations in the frequency domain [ 2 ] showed a n Qmaxas possible in the avalanche region, without a reoptinlum growth factor for small-signal oscillations ocduction of the electric field in the drift region below t h a t curringwhentheoperatingfrequency \vas about 20 required for velocity saturation when this charge moves percent higher than the resonance frequency. Therefore, throughthedriftregion. The motion of Qmax through \\-e take as a designcriterionthattheavalanchefrethe drift region results in an ac voltage amplitude about quency of a Read diode should be related to the drift Fig. 1. Diode and associatedcircuit.(a)Voltagedriven. running oscillator. (b) Free- c Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. JANUARY DEVICES, ?’R;\NSXCTIONStON ELECTRON IEEE 66 frequency of the diode by the following condition: f D s 1.2fR. (2) 1969 (2TfR)’ The small-signal constraint above, coupled with largesignal conditions imposed on the drift space to be discussed below, results in a tentative design for the Read diode n-hich is analyzed numerically in this paper. Thesquare of theavalanche regionresonancefrequency is proportional to the dc current density J d c r the carrierscatteringlimitedvelocity z J , andthepartial derivative of theaverageionizationrate B.4 forelectrons in the avalancheregion with respect to the average electric field EA.-4n approximation for the relationship is [ I ] , [21, [41 ( 2 q f ~= ) ~2B-4‘7’,lJdc/’~, (3) \\-here E is the dielectric constant, and BA’ the partial derivation of B A \\-it11 respectto E,. Thedcvoltage developed across the avalanche region is defined as I’dr..I = EALA, (4 and the dc voltage developed across the drift region similarly defined as L’dcD = EDLD, \\here L A and L O arethe effectiveIvidths alancheanddriftregions.l17iththedefinition frequency is ( 3 of theavof drift by ( l o ) , (8), and evaluated for theconditionsgiven ( 7 ) , is 2 x6x3x VdcDfD2/lidcA, (1 1) and with the condition onfR andfD given by ( 2 ) results in a relation between VdeA and I”dcD: VdcA 1 : 1.1VdcD. (12) This result is only qualitatively correct, sinceit is based on a combination of small-signalresultsfortheavalanche region, and large-signal constraints on the drift region. Furthermore, the factor 2 in (3) a n d ( l l ) , t h e factor 6 in (10) and ( l l ) , and the factor 3 in (9) a n d (11) are all approximate, and depend to some extent upondcbias,frequency,andoscillationamplitude. HoLvever, the qualitative result is important: the avalancheregionvoltage,for a Readdiodedesigned for efficient high-poxver oscillator operation, is on the order of the drift voltage and not negligible as suggested by Read. This condition reduces the theoretical efficiency by about a factor of 2 (15 percent instead of 30 percent). Xote that for Ge, GaAs, or other materials with nearly equal ionization rates, the factor 3 in (9) becomes a 1 and 1’dcAs+Vdc~, hence the theoretical efficiency is of the order of 23 percent. 111. LAKGE-SIGSAL CALKLATIOX Inthissection, we discussthepreciselarge-signal operatingcharacteristicsof a particularReaddiode oscillator, as obtained by a numerical calculation perfD = ?‘,1/2LD, (6) formed on a high-speed digital computer. Details of the computer program are discussed in the Appendix. 1 7 d c D can be expressed as BrieflJ-, theapproach is toobtainself-consistentnuVdcD = ED?‘al//’2fD. ( 7 ) mericalsolutionsfortheequationsdescribingcarrier transport, carrier generation, and space-charge balance For 50-percent voltage modulation ( T T a c ‘ v $ 1 7 d c D ) , Qnlas in one-dimensional a semiconductor structure. The from Gauss’ theorem is about ‘ v d ? D / 2 . The ac particle solutions describe the evolutionin time of the diode and current (\vhich for 100-percent current modulation itsassociatedresonantcircuit.Basically,theprogram equals the dc current) equals Q m a x f D ; therefore, the dc solves the follolving problem a t variousinstances of bias current densit)- for efficient operation of the drift time during a cycle of oscillation. Given the instantaspace is proportional to frequency: neous distribution of the hole and electron concentrations and terminal boundary conditions, how will the Jdc = € E D ~ D / ~ . (8) carriers move in time, i.e., how will the system, diode, For ionization rates appropriate for silicon [3], we find, and circuit time evolve? Given the carriers, the program for junctions \\-it11 breakdoum fieldsbetlveen 3 and determines the electric field by Poisson’s equation, and 4 X lo5 \‘:’cm, that the integralof the electron ionization knolving the electric field and carrier concentrations, it rate n-it11 distance is not unity, as is the case for equal obtainstheinstantaneous hole andelectronparticle rates, but about 3 : currents (including the dependence of carrier mobilities upon impurity concentration [ 7 ] and electric field BA LA ‘v 3. (9) strength [8], [9]). From the particle currents and net If n e assume for simplicity that B.4 varies as the sixth generationrate(includinggenerationandrecombinapower of E,, we obtain tion by field-dependent impact ionization [3] and carrier concentration-dependent single-level recombination Bn’ ‘v 6BA/EA centers [ l o ] , [ l l ] ) ,it obtains the time derivative of the or, using (9) and (4), carrierconcentrationsfromthecontinuityequations. From the time derivativesof the carrier concentrations, B.4’ ‘V 6 X 3;’ V d c . d . (10) the program computes the carrier distributions an inEquation (3), theavalancheresonancerelation, n-hen stant in time later, and repeats the cycle. Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. SCHARFETTER AND OSCILLATOR DIODE GUMMEL: READ SILICON The diode operating characteristics are calculated for the diode imbedded in the circuit shoivn schematically in Fig. 1(a), and (b). The variation of diode negative conductance, susceptance, average voltage, and microwave power-generating efficiency withfrequency,dc current density, and ac voltage amplitude was obtained with the circuit shown in Fig. l(a). This method made it possible tocontroldirectlythefrequencyandac voltage amplitude in the calculation and was employed to economize on computer time. Essentially the same results were obtained with the circuitsholvn in Fig. l ( b ) , if theadmittance of the loadinductance L and load conductance G Lvere properlychosen.Thecoupiing capacitor C served only to isolate the dc and ac portions of the circuit so that the dc current density could be directly controlled. The doping profile for the diode studied in detail is shown inFig. 2 . The corresponding steady-state electric field versusdistanceforthisdiode is shownin Fig. 3. T o simplifythesignconventionon field and current density, thefigures representa n-v-n-9 structure with the substrate on the left and the diffused junction on the right. Note that the donor densityin the v region is not negligible, but causes a continuous fall-off of electric field out to the substrateregion. This doping profile evolved in trial runs to eliminate generation by impact ionization asthecharge pulse of electrons passed through this region, while still allowing as large an average electric field in thedrift region as possible(see Fig. 12 (d) I\-hich illustrates this effect). Inagivenrunwith fixed dccurrentdensity,ac voltage amplitude,andfrequency,theprogram \\-as continued until the limit cycle was obtained. The terminal currentandvoltagewaveforms,during a limit cycle, were analyzedtoinvestigatethevariation of diode characteristics at various values of oscillation amplitude. Thethreefundamentaldiodecharacteristics investigated arenegativeconductance,susceptance, and averagevoltage.Resultsare sho\\-nin Figs. 4, 5 , and 6. Thediode’slarge-signalnegativeconductance and susceptance values are effective values at the voltage generator frequency obtained by Fourier analysis of the terminal current and voltage waveforms. Detailed results are shown for only one current density because of the great number of lengthy computer runs required to characterizecompletelythelarge-signalbehavior. However, calculations made at other current densities (100 to 1000 X/cmZ) indicate that the results are qualitatively similar to the detailed results presented for a current density of 200 A/cm2. The magnitude of negative conductance increases up to about1000 A/cm2, but the efficiencies obtainedare less t h a n a t 200 A,/cm2. Thediode’ssusceptance is predominatelyduetothe diode’sspace-chargecapacitance,andthereforedecreased with increasing current due to widening of the space-charge width. The average voltage increases with currentdensity,butbecause of therapidincrease of ionization rates with electric field, the average voltage 67 decreases \\-it11 ac voltage amplitude \\-hen the average current is held constant.Large-signalresultsatlarge currentsare moredifficult toobtain \\-it11 thecircuit shown in Fig. l(a)(voltagegenerator) because parametric [12] effects occur a t large dc current densities. Thesusceptanceshown in Fig. 5 is expressed asa “capacitance” and plotted in Fig. 7. The value is only slightly less than the depletion layer capacitance (1.9 X lop9 F/cm*), n-ith thelargestdeviation at the lower frequencies. The ratio of susceptance to conductance Q is shown in Fig. 8, and the power-generating efficiencies are plotted in Fig, 9. The conductance and susceptance values are sho1j.n as an admittance plotin Fig. 10. Results for frequencies between 6 and13.4 GHz andacvoltageamplitudes between 0 and 38 voltsaresho\\-n for adccurrent densit>- of 200 -\.‘cm2. Alsoshon-n areconstant efficienc\- contours. The plot indicates, for example, that 10 percent efficiency is obtained a t a frequency of 11.4 GHz \\-it11 a voltage amplitude of 30 volts, \\bile only 20 volts is required a t 8.4 GHz. I t \\-as found that the efficiency increasesu-ithacvoltageuntiltheelectric field modulation is so large that, over the negative part of the voltage cycle, the field dropped so low that the carrier velocity was less than the saturated value. This occurs a t considerably lolver voltage amplitude for the lower frequencies (less than 7 C H z ) , and made it very difficult to obtain steady-state solutions for thesefrequencies.Also,notetherapidincrease in negative Q (ratio of susceptance to conductance) \\-ith ac voltage amplitude. This would cause the efficiency to turn over and decrease with further increases in ac voltage amplitude for deviceswithseriesleadandcontactresistance, since the diode shunt negative resistance’ is reduced by a factor of Q 2 in converting from a parallel to a series equivalent circuit. Detailed “snap shots” of the electric field, and hole andelectronconcentrationasfunctions of distance during one cycle of steady-state oscillation are shon-n in Fig. 11. The “snap shots” are shon-n a t approximately one-fourthcycleintervalsinFig. l l ( a ) through(d). T h e figures are selected frames from a computer-made movie [ 6 ] . .I phase plot of the terminal current I and voltage 1- of the oscillation is included in the figures. Points to note are the follotving. 1)Thegeneration of pulses of holes andelectrons begins\\-here thevoltage is amaximum;one-fourth cycle later, the charge pulses are fully formed and begin drifting into their respective drift spaces. 2) T h e holes disappear quickl!. fromtheactiveregion while the electrons driftfor approximately one-half cycle and constitute positive particle current while the ac voltage is negative. The calculationincludes all losses that originate between the electricalcontacts, i.e., substrateandavalanche space-chargeresistances. Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. 68 IEEETRXKSA\CTIONS O S ELECTROSDE\.ICES, I60 J-ASUARY 1969 13.4 12.4 140 3.60 3.2- E ,o VI 3 c 11.4 1201 2.8- -9a 2.42.0lA ----0.4- Fig. 3 . Staticelectric field and particlecurrent profile versus distance i n dc stead?--state Read diode 7&1. L 0 I 10 I 20 I 30 I I 40 '50 IO A 0 & DISTANCEINMICRONS A C VOLTAGE AMPLITUDE Fig. 1. Diodenegativeconductance(mhos/cm*) as a function of acvoltageamplitude for ~ a r i o u sfixed frequencies.Current: density 200 .1/crn2. 30 40 AC VOLTAGEAMPLITUDE Fig. 5 . Diodesusceptance(mhos/cm*)as a function of acvoltage amplitudeforvarious fixed frequencies.Currentdensity: 200 A/cm2. I 6 20 0 IO I I I 20 30 40 A C VOLTAGE AMPLITUDE I 50 I 0 Fig. 6 . Diode average voltage as a function of ac voltage amplitude for various fixed frequencies. Current density: 200 A/cm*. Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. 69 SCHARFETTER AND GUMMEL: SILICON READ DIODE OSCILLATOR 17.5I 850- 15.01,825- 8 W g a 12.5 - > 1.800- V z t V lg 10.0 a a a 1.775- IL IL W u 1,750- - 7.5 5.02.5- 1.725- -T - 2 I I 0 IO 30 I I / I I 20 I 01 0 50 40 AC VOLTAGE A M P L I T U D E IO I 20 I 30 I 40 I 50 1 PC VOLTAGE A M P L I T U D E Fig. 7. Diodesusceptance expressed asanequivalentcapacitanceFig. (depletion layer capacitance= 1.9X F/cm2)versusacvoltagevarious amplitude for various fixed frequencies.Currentdensity: 200 A/cm2. 9. Power-generatingefficiellcyversusacvoltageamplitude fixed frequencies. Current density: 200 A/cm2. 160 I50 140 I30 I20 I IO 100 W 90 Q 80 x4c 3 2 0 70 60 50 40 30 20 01 I 0 IO I I I 20 30 40 I e AC VOLTAGE A M P L I T U D E Fig. 8. Diode Q (ratio of conductancetosusceptance)versusac voltage amplitude for various fixed frequencies. Current density: 200 A/cm2. I -20 -15 IO 0 I -10 -5 0 CONDUCTANCE Fig. 10. Diodeadmittance(susceptanceversusconductance)asa function of frequency and ac voltage amplitude, and resultant efficiency indicated. Current density: 200 .4/cm2. Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. for 70 JEEE TRXKSXCTIONSONELECTRON DEVICES, JANUARY 1969 -5.0 Y) - 4.0 * I E \ Y) - 3.0 2 -a> a1 A - 2.0 u a + u 0.4 W J W - 1.0 Lb 0. 0 w L l 1 2 3 0. I 4 5 6 7 8 3 IO DISTANCE IN MICRONS DISTANCE MICRONS IN (C) -5.0 -0.6 v, w a U - 0.4 - 4.0n0 * .5 I 1 2 3 4 . 5 E - 3.0 - 3.0 v, -> -0 0 0 J - 2.0 w W - -20‘ 9 0 - K + u + u W - 1.0 I I 5 6 7 8 * - E - 0. 0 - 4.0: - / E -5.0 W J W - 1.0 W L0. L0. IO 9 DISTANCE IN MICRONS J 9 DISTANCE IN IO MICRONS (b) (a Fig. 11. Solutions of hole and electron concentrations, electric field, and terminal current and voltage (values indicatedby on phase plot) at various points in timefor the diodeoperating at a frequency of 12.1 GHz, current density of 200 A/cm*, and efficiency of 12 percent. 3) For the next one-fourth cycle, the remnants of the electron charge pulse are swept out of the picture as the voltage again approaches its maximum value, 4) The displacement current is quite large and has anappreciableswingintotheconventionalfor\\-ard direction, n-hile the terminal voltage always remains in the conventional reverse polarity. 5) The behavior is as predicted by Read, except that forsilicon diodesanextendedavalancheregionisrequired relative to the drift region to obtain the magnitude of charge pulse necessary for sufficient modulation of diode voltage and particle current for efficient oscillations. Efficient oscillations were also obtained a t higher dc current densities by trial and error choices for the load con’ductance G and load inductance L. “Snap shots” of the oscillation at a dc current density of 1000 A/cm2 are shown in Fig. 12(a) through (d). The largest efficiency obtained for this bias (9 percent) occurred at an operating frequency of about 13.4 GHz. Kote that the carrier concentration scales differby a factor of 5 between Figs. 11 and 12, as do the dc current densities. However, the large-chargepulse, for the 1000 A/cm2case,moving through the drift space results in a significant value of peak field near the substrate just as the electrons approach this region. I f this peak field is too large, genera- Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. 71 SCHAKFETTER AND GUhIMEL: SILICOK READ DIODE OSCILLATOR -5.0 1 5.0 - 4.0 in 1; . 0 - P -E v) 4.0 * -3.0 v) 5 E - 3.0- -> r n E -1 - 2.0 w a V 2.0 - Y I v) r 3.0 - w U a V 2 0- I- .u W J -1.0 0. I 0 , 1 O2 3L i 4 DISTANCE 5 IN 6 7 8 k 0. 9 IO 1.0 1 0 1 2 3 4 5 6 7 8 DISTANCE IN MICRONS MICRONS (a) (C) -5.0 -4.0 m -Iz* .5 v) -3.0 3 0 > I 0 J w -2.0 U + u W J - 1.0 W DISTANCEINMICRONS DISTANCE IN MICRONS (a) (b) I;ig. 12. Solutions of hole and electron concentrations, electric field, and terminal current and voltage (values indicated b\. 0 on phase plot) at various points in time for the diode operating a t a frequency of 13.4 GHz, current density of 1000 A/cmZ, and efficiency of 9 percent. tion byimpactionizationoccurs,andthesubsequent drift of a hole charge pulse toward the p region would constitutepositiveparticlecurrentnhilethevoltage was alsopositive,averyunfavorablephaserelation. Thiseffect was worse in an earlier doping profile (not shon.n), and was eliminated by altering the profile to provide a continuous fall-off of electric field out to the substrate (see Fig. 3). The nearly ideal (classical Read [l phase relations obtained between diode voltage and particle current is illustrated in Fig. 13. In this figure, we plot the waveforms in time of the particle current and terminal voltage for a n ac steady-state solution at 11.4 GHz and dc current density of 200 A/cm2. The particle current was 1) obtained by subtracting the displacement current from the total current. At the largest oscillation amplitude, thediode“capacitance”changesparametrically[12] and makes an accurate separation of the current componentsdifficult. The result shown inFig. 13 is for a voltageamplitude of about 17 volts.Holyever,the waveforms for larger amplitudes are similar. The particle current is not a square wave during the half cycle over which the voltage is negative, but has a hump resultingfromtheextracomponent of particlecurrent which flows until the holes that were generated along with the electrons are swept into the p layer. For large amplitudes, the hump becomesmorepronouncedand the particle current bottoms [all carriers are swept out Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. 72 IEEE TRASSACTIOSS O S ELECTRONDEVICES,JANUARY 1969 APPENDIX SEMICONDUCTOR DEVICEA S A L T S I S COMPUTER PROGRAM 1 INTRODUCTIOX ThisAppendixconsists of threesections.Section I describes the device physics included in the large-signal calculationandoutlinesthesolutionprocedureswithout detail.SectionI1describesindetailthetimeadvancement techniques and the mathematical formulation employed in the computer calculation. Section I11 is a detailed discussion of the matrix inversion subroutine developed specifically for the time evolution computer program. I. ~ I A T H E M A T I C A L~ I O D E L A . Charge Transport Equations 0 TIME Fig. 13. Diode voltageandparticlecurrentversustime cycle of steady-state oscillation. for one as shown in Fig. 11 ( a ) ] over a n appreciable part of the cycle. The resultsof the calculation presented in this section have shown that a properly designed silicon Read diode oscillator is operable,with efficiencies from 9 to 18 percent,overanorder of magnitudevariation in dc current.Thefrequency of operation,amplitudes of voltage and current oscillations, order of magnitude of dc bias, and efficiency are in fairagreementwiththe prediction of therelativelysimpledesigntheorypresented in Section 11. The distribution and motion of carriers Ivithin a onedimensional semiconductor device structure can be obtainedbysolvingthreebasicequations:1)the continuity equation forholes, 2 ) , the continuity equation for electrons, and 3) Poisson’s equation: 1 d.T, dP -df= g - - - , q 8% dl1 1 dJ,, dt q _-g + - -> dm and \\here IV. COXCLVSIOSS Operating characteristics of a silicon Read diode \\-ere obtained from large-signal computer calculations of the evolution in time of the diode-resonant-circuit system. The values of ionization rates [ 3 ] and carrier velocities [8],[9] appropriate for silicon were used in the calculation.Self-consistentsolutionswereobtained for the equationsdescribingcarriertransport,carriergeneration,andspace-chargebalancen-ithinthediode,and n-hich simultaneously satisfied the boundary condition imposed at the diode metallic contacts by the resonant circuit. Detailedsolutions were presented of the hole and electron concentrations, electric field, and terminal current and voltage at various pointsin time during ac steady-state cycles of oscillation. The largest efficiency obtained (18percent) is in good agreementwiththe approximatecalculation,whichincludestheappreciablewidth of the avalanche region requiredfor silicon devices.Efficientoperation ( 9 to 18 percent) was obtained over a range of dc current density (100 to 1000 A/cm2) and frequency (6.4 to 13.4 GHz). Boundary conditions are imposed at the contacts by introducingtheappropriaterestrictionsin(13)through (1 7 ) . For example, current boundary conditions for a p-n deviceareintroducedbyrequiringthat(17) is equal to the terminal current density a t t h e n contact and that (16) is equal to the terminal current density at the ‘ p contact. \-oltage boundaryconditionsareimposed by requiring that the integral of E ( x , t ) over the intervalbetweenthetwocontactsequalsthetotal voltage. I n addition, the electric field at the two end metallic contacts is assumed to be zero. Initial values for the hole and electron densities are either furnished by a previous run or are given by the quiescent zero bias solution, i.e., the solution when Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. 73 OR E READ SILICON GUhlMEL: SCHARFETTER AND In general, the functions g, p,, p n vary at each point in the device according to the value of n , p , and E a t that point. B. Carrier Generation-Recombination Expressions Thecarriergenerationterm g iscomposed of two components: 1) carriergenerationandrecombination through defects, and 2 ) impact or avalanche ionization. Hole electron generation and recombination through defects are represented by a Shockley-Read-Hall (SRH) single-levelmodelwhichcharacterizesdefects with neutral and single-charge states. The generationrecombinationratethrough single-level a center is given by p n - ni2 gd + nd + + PI) = -- rpo(n Tno(P (19) Theimpactionizationratesarestronglydependent upon the electric field intensity and the hole and electroncurrentdensities.Thesegenerationtermsare given by where an(E) and a,(E) are given by the relations an(E) = 2.25 X lo7 exp (-3.2 X 106,/E) (21) a,(E) = 3.80 X lo6 exp (-1.75 (22) X 106/E). Ax q = -(P(X) - n(n.) E Holes Electrons s 480 4X1Ol6 s A F P(NP,(M) J,(M) = E(M) J,(M) = 12(-I- l)Pn(M) E ( M )_ _ _ _ - ~ (1.0 - esp (-E(M)Ax)) (1.O - exp (- E ( M )Ax)) + t --- (1.0 - esp ( E ( M ) A $ Theseequationsprovidenumericallystableestimates of the current density under all conditions. If the intermesh point voltage is small, these equations approach thestandard differencerelations;whereaswhenthe voltage change is large, they approach the drift current density at either mesh point N or N+1. + (E/BI2 PO (27) -ITA (X)) where the Mth mesh point is located midway between the major mesh points N + 1 and N . I t is customary to employ next the standard difference approximations in the current density expressions (16) and (17) and substitute these results in ( 2 5 ) and (26). However, it can beshon-n thatthisprocedureleadstonumericalinstabilitywheneverthevoltagechangebetween mesh points exceeds 2kT/q. Rather, (16) and (17) are treated as differential equations in p and n with J,,, J,, p,, pn, and E assumed constant between mesh points. The solution of these differential equations then relates J , and J , to the other variables: C. Mobility Expressions I t is necessary to include in the analysis the variations of mobility ( p ) withtheelectric field ( E ) andthe ionized impurities density ( N o ) . Thetheoreticalmobility is approximatedbythe following expression : + -I-o(A') - R 81 6.1X103 1 . 6 2.5xlO4 1400 3X1Ol6 350 3.5X103 8 . 8 7.4X1O3. (24) D. Solution Procedures Because of thenonlinearities in theequationsdescribing the hole, electron, and field distributions, obtaining a transient or even a steady-state dc solution poses a very difficult numerical problem. The structure to be analyzed is first subdivided into a number of small cells. Theequationsarethennormalizedtoreduce redundant coefficient calculations, and standard difference approximations are used to approximate the spatial derivatives in Poisson'sequationandthecon- E . Definition of Symbols hole and electron concentrations, cm-3 electric field intensity, V/cm electronic charge, 1.602 X coulombs hole and electron current densities, A/cm2 hole and electron generation-recombination rates, carriers/'cm3.s impactionizationgenerationrate for holes and electrons, carriers/cm3.s hole and electron generation-recombinationratesthrough a single-level trappingcenter,carriers/cm3 .s dielectric constant Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. I E E E TRANS.ACTIONS ON ELECTROXDEVICES,JA4NUARY 74 Boltzmann constant, 1.38044X J /OK form, b y the closed-form expression conventional 1969 (33). Written in more absolute temperature, OK total ionized impurityconcentration, ~ m - ~ (36) concentration O f donor and acceptorEquation (36) is thevectorequation we solveforeach impurity atoms, c 1 r 3 increment t intime. LYhen n e havesolved it, \ve and PP,P n c m 2 / v ' s incrementthevector yo bythesolution 6y. To geta low-doping 'le PPO i n better intuitive understanding of what we are doing, let silicon, (480 c m 2 / V . s ) us move the term containing 31t in (36) on the rightlow-field low-dopingelectronmobilityhandside: in silicon, (1400 cm2/V.s) ionization coefficients for holes and (3 7) electrons, carriers/carrier cm hole and electron concentrations in the conduction band \vhen the Fermi level -At titne 0 , the slope 5 isf(yo). At time t , i t is approxicoincidesLvith theenergy level of themately fo+-Tf6y. Theratio of the finitedifferences single-level trap. 1I. VECTOR-iVIATRIX F o R h f C L A T I o s A . T i m eA d v a n c e m e n t We represent the hole and electron concentrations by the two-component vector y . The system of equations then reads i.e.,thetimederivative of y equalssomenonlinear functions f of y . W'e denote y a t t h e beginning of the time interval under stud!- by y o , and denote the deviations of this value as time increases by 6 y ( t ) : y ( 0 = yo + 6y(t). (3 1) (1 - M t ) S y = f*t. For small deviations 6 y , we expand f ( y ) as 6 i = f(y0) + M6y. (32) is the matrix d f / d y and represents an integro-differentia1 operator. Kow, if 51,f, and 6y were ordinary numbers, the solution to (32) would be 6y(t) = (eiMt - l)f/M 6 y ( t ) / t corresponds slope the to ofstraight the line 0 and t . \That we aredoing - is through - thepoints propagating the system according to the time derivative a t some advanced time t a e t / 2 , rather than at the initial time. The term in 6y on the right-hand side of (37) represents a feedback that gives the system stability. I t is worth noting that the expansion off with respect to y (the derivative matrix M ) is correct to first order, but with regard to timeit is correct to second order in t. The method of propagation (36) time evolves the system accurately for small time steps t. However, if we are near an equilibrium state and are interested not in thetimeevolution,butonly in theequilibriumsolut of infinitycan he taken n-ith full tion,timesteps feedback, i.e., B . The Derivative dlatrix We define the vector y as having components of hole concentration, electron concentration, and electricfield : (33) or 6y(t) =f.t(l + M t / 2 + ( M t ) ? / 6 + . . . ). (34) and the right-hand side of the continuity equations as Note that if the physical situation is one in which \ve the vector f: are near a stable equilibrium state, then the eigenvalues of the matrix M must be negative so that as timegoes to (39) infinity, 6y reaches a finite value. For small but finite time steps, we introduce the inThe vector equation we solve for each time increment verse of the first two terms of (34) to obtain our solution t (36) is to (30) a s L. where Aft represents changes in fs at the terminals and where we define the derivative matrix Theoperator A1 in thedenominatorroughlycorresponds to integration and as t goes to infinity, we reach an asymptotic value within a factor of 2 of that given Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. 75 ERREAD SILICOX GUMMEL: SCHARFETTER AND The matrix elements of X are defined by the expanded form of (40) shown below: + ( F L ) M* y3.tr-l + (FR),?r* = X L ( 3 , -V) * 6 H ( S - 1 ) + X L ( 4 , 17') * 6E('q7 - I ) + + (44) X R ( 3 , N ) * 6H(-V f 1 ) + X R ( 4 , JY)* 6E(.Y + 1) + XM(3, S ) y a = ~ (RH)N (42) where (RH1.v and 2(fo).v * 6H(.\') + X M ( 4 , -I-*)6 E ( S ) + I'ector equation (42) is solved by the computer program FL(2, S ) * 6 F ( M - 1) FR(2, -Y)* 6 F ( M ) subroutine described in Section 111. T h e 2x2 matrices = RH(2, -\.) S L , X M , and X R are obtained by taking the partial derivatives off.^ with respect to H@L andELE a t mesh \\here N ranges from 1 to L R , and points N - 1 , N , and N+1, respectively, while the two component vectors FL and FR are obtained by taking 6 F ( M ) - 6F(M - 1) = (6H(.l-) - 6 E ( N ) )* D Y ( S ) the partial derivatives off.\. with respect to FIELD at where mesh points IZI and 21- 1, respectivel!,. 111. ~ I A T R I ISVERSIO~X SUBROUTIXE Dl'(.\.) = (DX(M) + DX(M - 1))/2. (45) This section describesin detail the method for solving Inputquantitiesandtheirrequiredrangeare as thediscretizedcontinuit).andspace-chargeequations follo\vs : for the quantities 6 H ( N ) ,6 E ( N ) ,and s F ( 3 I ) .The quanDAY.~r M = 0 , LR tities 6H and 6E are the changesin the hole and electron S = 2, LR [XL], concentrations, while 6 F is a normalized electric field; 6F is in units of carrier concentration. The boundar!. [XM].h, S = 1 , L R conditionshavebeenincludedintheequations,i.e., [ X R ] # S = 1, LM (LM = L R - 1) the particle currents at the left contact enter into the ( R H ) N -1-= 1, LR right-hand side R H ( 1 ) of the first continuity equation, and the particle currents at the right contact enter into -Y= 2 , L R (FL)K theright-handside R H ( L R ) of thelastcontinuity (FR)x S = 1 , LM. equation.Sincethetotalcurrent atthecontacts is known, continuityequations 1 and L R containthree Output quantities and their range are as follows: houndary conditions. The fourth boundary condition is 6H.v = Y(1, S ) S = 1, L R that the electricfield at the contactis zero. Since overall space-chargeneutrality is builtintotheequations, a 6E.v = Y ( 2 ,S ) .\- = 1 , LR field of zero a t one contact forces a field of zero a t t h e 6F.11 = Y ( 3 , M ) M = 1, LM other contact. The method forsolving thediscretized (note 6F(O) and 6 F ( L R ) are zero). continuity equations will bedescribed for the case of terminal-current boundar!. conditions.ThemodificaB . Dettrils of Inversion tions to this method \\.hen boundary conditions are in terms of terminal voltage a mixture of voltage \Ye represent the output quantities6H and 6E by the - or - and current Ivill be described in Section 111-C of thisvector y as .Appendix. -4. Input-Output \Ye describe a computer subroutine which solves the following three equations: X L ( 1 ,S ) and similarly input quantities RH,F L , FR as * 6 H ( S - 1) + X L ( 2 , -T) * 6E(&17 - 1) + + (43) X R ( 1 , X) * 6H(,1' + 1) + X R ( 2 , X) * 6 E ( S + 1) + X M ( 1 , &4-) * 6H(L3-) + X M ( 2 , S ) * 6E(A7) FL(1, S ) * 6F(M - 1) + FR(1, -I-*)6 F ( M ) The input quantities X L , X M , XR are represented by Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. IEEE TRANSACTIONSONELECTRONDEVICES,JANUARY 76 [XLI, = [XX LL (( 13 ,, 1, X) XL(2, S ) AT) XL(4, X ) 1969 Note that (48) can be written as - ( U F ) L * 6F(L) and (45) as ~ F (L 1) = ~ F ( L ) J ~ H ( L-) ~ E ( L )*} D Y ( L ) ; (53) therefore, once we have [ U M ] , ( U F ) , and ( 2 ) a t mesh points 1 to L R , we can n-ork backwards using (52) and (53) to find and ~ F L - Ifor L = LR . 1. The computation procedure is as follows. 1) Compute starting values from (51). 2) Computevaluesattheremaining mesh points ( 2 , L R ) using (50). We assume that \ye can remove the ( Y ) . Y - ~ term in (47) 3) Use ( 5 2 ) and(53),starting at mesh points L R and write the resulting equation at mesh point N-1 as (note ( y ) L ~ + 1and 6F(LR) are both zero) and working [CMIN-l * ( Y ) N - l [URIN-l * (y1.v (48) through mesh point 1 to obtain the values for 6H, 6E, and 6F. (UF)N-l * 6 F ( M - 1) = @)x-1. where + + Recursion relationships for [ V M J , [ UR],( U F ) , and (2)are obtained as follows. Equations (47) and (48) arecombinedtoeliminate (y).v-l terms as { [ x M ]-. [L'FL].v ~ - [ ~ ] *x [ ~ ' ~ ] . v -*1(y1.v ] [ X R ] s* ( ~ 1 ~ 1 (~ F 1N M (FL).,r) * - p 1 . v * (CF),V-l * 6 F ( M - 1) = ( R H ) . y - T2v* (2)s-l p1.v = [XLIX * [ z - M ] N I 1 . + + + the desired boundary conditions could be incorporated intotheinput coefficients. The terminal currents enteredintotheright-handside (RH),v at the contacts ( N = 1) and ( N = L R ) . RIore general boundary condifiF(m) tionscanbeincorporatedbyusinganeconomical (49) superposition process. Withthetimeadvancement processdescribed in Section 11, we n-ant to solve the equations Theterm in 6F(m- 1) can be eliminatedin (49) by (45), and therefore by comparing (49)Irith(48)evaluated at mesh point N , n-e get: [ m ] , v = [XRIY [IW].v .)I:( (Z'7F)1= = [XM]s - [L-FL],v- C. General Boundary Conditions In the discussion thus far, i t has been assumed t h a t [T]X * [z'R].v-l [f - "21 * (Y) =f, (54) where t is thestep intimeand M is thederivative matrix. If the terminal currents are changing with time, then f is fo, the right-hand side of the continuity equation at t =0, plusone-half of thechangeinterminal current (the extra term will appear in only two equations, mesh points l and L R ) . Let us multiply (54) by 2 and write it as where A j represents the change in terminal currents in the time t. Equation ( 5 5 ) is identical to (46) and could be solved by the method discussed previously in Section 111-A of this Appendixif the changein terminal currents (Af terms) are known. What is alwaysknown a t t h e terminals is one of three possibilities: 1) the terminal current as a function of time is an independent variable; Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply. 77 R E READ SILICON GU1\’lMEL: SCHXRFETTER AND 2) the terminal voltage as a function of timeisan independent variable; or 3) an equation relating terminal current to terminal voltage is known. The general solution to (55) for Af arbitrary can be obtained numerically using superposition. Letus assume that our final solution will be the sum of two solutions: (Y) = (Yo) + (YJ. (36) These solutions have the following interpretation: (yo) represents the solution for Af = 0 , that is, for no change in terminal current over the time step t ; (yc) representstheadditionalsolution in terminal current. T o see this more clearly, let for a change u s n-rite ( 5 5 ) as [ M ] N ( y ) = (RHOIN + (RHc).V (ji) where the quantity 2 l t + d i in (55) has been n-ritten as [ I l d ] .in ~ ( S i ) and 2fo as (RHo).v and Af a s (RHC).y. Equation (5i) has (RHc).vzero everywhere, except a t mesh points 1 and LR.Therefore, u-e solve ( S i ) using the method described in Section I I I - L lof this -Appendix n-ith (RH).y = (RHO).,.. This yields solutions (yo).,-. S e x t , we solve ( 5 7 ) with (RH).\T = (RHc).\-;this yields solution (yc).,-, Since Ive donotknowthechangeinterminal current A J , we obtain solutions (yc).v for a unit change in terminal current. Lye then integrate the electric field component of (yc).y toobtainthechange in terminal voltage per change in terminal current Av,. If the integral of theelectric field component of (yo).\. is called AI’,, then the change in terminal voltage A I 7 is related to the change in terminal current A J b\AV = AVO + Avc * AJ. (58) REFERENCES [l] \V. T. Read, “A proposedhigh-frequencynegative-resistance diode,” Bell Sys. Tech. J . , vol. 37, pp. 401-466, March 1958. H.K.Gummeland D. L.Scharfetter.”Avalanche region of IAIPXTTdiodes,” Bell Sys. Tech. J . , vol. 45,pp. 1797-1827, December 1966. C. A. Lee, R . X. Logan, I<. L. Batdorf, J. J . Kleimack, and ii-. iViegmann, “Ionization rates of holes and electrons in silicon,” Phys. Rev., vol. 137, pp. i\761-X773, May 1964. M. Gilden and 1’1.B. Hines,“Electronictuning effects i n the readmicrowaveavalanchediode,” I E E E T r a n s . ElectronDevices, vol. ED-13, pp. 169-175, January 1966. T. Misawa,“Negativeresistance in p n junctionsunder avnlanchebreakdownconditlons,parts I and11,” I E E ET v a n s . Electron Devices, vol. ED-13, pp, 137-151, January 1966. D. L. Scharfetter and H. K. Gummel, “Design of Readdiode oscillators,”presented at the 1966 IEEE Solid-stateDevice Research,Conf.,paper I11 b-4. J. C. Irvln, “Resistivity of bulk silicon and of diffused layers in silicon,” Bell Sys. Tech. J., vol. 41, pp. 387-410, hlarch 1962. .A. C. Prior, ..‘‘Field dependence of carrier mobility in sllicon and germanium,' J . Phys. Chenz. Solids, vol. 1 2 , pp. 175-180, January 1960. T. E. Seidel and D. I,. Scharfetter, “Dependenceof hol: velocity upon electric field ;tnd hole density for P-type silicon, J . Phys. Chem. Solids, vol. 28, pp. 2563-2574, 1967. \V. Shockley and I\’. T. Read, Jr., “Statistics of the recombination of hole and electron,’’ Phys.Rev., vol.87,pp. 835-842, September 1952. R. X. Hall, “Electron-hole recombination i n germanium,” Phys. Rev., vol. 87, p. 387, July 1952. B. C.DeLoach. Tr.. andR. L. Tohnston.“.\valanche transittime microwaveoscillators an& amplifiers,” I E E ET r a n s . Electron Devices, vol. ED-13, pp. 181-186, January 1966. Authorized licensed use limited to: University of North Carolina at Chapel Hill. Downloaded on April 09,2010 at 21:39:33 UTC from IEEE Xplore. Restrictions apply.