ion the mobility is in agreement with the alues found for the MOS transistor. In the aturation region the decrease in mobility s consistent with the reduction in mobility ith increased electric field in the hannel.

The 32-bit circular CCD was operated in a Charge transfer mode in order to measure the ast interface state density per unit enerY, NEs, by the double-pulse method [35]. typically potentials on the three phase lectrodes were varied between 0 V and +15 v ith respect to a grounded substrate at a requency of 125 kHz. The input and output ates were held at +15 V when injecting or xtracting charge from the channel; otherise they were held at zero bias. In addiion, the input diode was slightly forward iased with respect to the input gate when he latter was at + 15 V, V = + 15 V, and ref = + 17 V. Using the double-pulse methd, a pair of "ones" followed by N

ps

zero

zeros

This quantity is related to the fast state density per unit energy, Ngs, by [35]

where k is Boltzmann's constant (eV/k), T is the temperature (K), N is the unattenuat

sig

ed signal charge density (cm-2), and F is the fraction of time available from each period during which charge can be transferred. This fraction can be determined graphically as shown in figure 18. For the three-phase CCD used here one would anticipate F≈ 1/3. The value determined experimentally, F = 0.29, is not unreasonable; the exact value depends on the details of the phase voltage rise and fall times and the interelectrode gap fringing fields. The value NFS = 1.2 x 1010 (cm2 eV)-1 follows

The power of the circular CCD configuration is illustrated in figure 19 where the data from figures 17 and 18 are replotted. The circular CCD allows the accumulation of a very large number of transfers which allows the pulses to decay to detectable levels. In a linear CCD, the number of transfers is fixed by the number of bits and cannot be changed. In addition, the double-pulse method establishes a number for N, which is FS more difficult to obtain from high-frequency MOS C-V characteristics.

(I. Lagnado and M. G. Buehler)

The factor F is not explicitly included in reference [35] but is implicit in the derivation.

Naval Electronics Laboratory Center, San Diego, California 92152.

7.1. Cleave-and-Stain Measurements

An attempt was made to stain the regions of the five specimens which had been anglelapped and probed by a spreading resistance probe as reported previously (NBS Spec. Publ. 400-8, pp. 32-33). By observing the stain, a measurement [36] was to be made of epitaxial thickness on each of the five slices. It was found, however, that the bevel angle, rather than having a welldefined apex, was rounded owing to preferential polishing at the apex during angle lapping. This would have caused the spreading resistance measurement and the lap-and-stain measurement, had it been made, to give erroneously large values of epitaxial thickness. An alternative approach was taken. The segment from each slice on which spreading resistance and step relaxation measurements had been made was cleaved at a 90 deg angle, stained, and photographed using a scanning electron microscope (SEM). The epitaxial thickness was then determined from the SEM photographs. The results of these measurements together with previously reported thickness determinations on these wafer segments are given in table 4. Although the cleave-and-stain measurements appear to be in somewhat better agreement with the step

relaxation measurements than the spreading resistance measurements, the agreement is still only fair. (J. R. Devaney and R. L. Mattis

7.2. Metal-Photoresist-Semiconductor

Capacitors

In order to apply the principles of the MCS capacitance methods (NBS Spec. Publ. 400-4, pp. 51-53) for epitaxial thickness measurement in a manner that is non-destructive and involves no high temperature processes and resultant impurity redistribution such as occurs during thermal oxidation, an effort was directed toward the use of photoresist as the dielectric in place of the oxide and, further, toward use of a photolithographed metal pattern which would provide a known and uniform device area. The specimen preparation which has been developed involves (1) spin-on application of negative photoresist to a clean epitaxial wafer, (2) bake to form a dielectric layer 0.3 to 0.7 um thick, (3) evaporation of aluminum over the dielectric layer, (4) spin-on application of positive photoresist, (5) bake of the positive photoresist and its subsequent exposure through the metal mask of test pattern NBS(see sec. 6.1.), (6) development of the positive photoresist, and (7) etching away the aluminum which is not a part of the desired metal pattern. The result is an arra of metal dots which form metal-photoresistsemiconductor [M(PR)S] capacitors analogous to the MOS capacitors which employ oxide as the dielectric. Epitaxial specimen 2213 was prepared by the above process and the transient capacitance characteristics of five M(PR)S capacitors along a slice diameter were recorded. The resulting values of layer thickness ranged from 5.24 to 6.04 m These values are consistent with the layer thicknesses of other epitaxial wafers from the same lot (see wafers 2203 and 2204 in table 4).

Improvements are still needed in the processing by which M(PR)S specimens are prepared in order to obtain thinner and more reproducible dielectric layers, but the feasibility of using such structures for thickness measurement has been demonstrated.

(J. Krawczyk and R. L. Mattis

The constant level of spreading resistance characterizing the substrate was not reached abruptly in this wafer making the thickness determination obscure.

3.1. Flying-Spot Scanner Development

Several additions were made to the optical lying-spot scanner in order to extend its isefulness. A superheterodyne receiver was idded to allow one to obtain the scanned response of structures to 0.633 um light moduated at 385 and 770 MHz, mode-beat frequen:ies internally generated within the particlar laser used. The scanner has been operited also with 1.15 um light; operation can e shifted from 0.633 um to 1.15 um and back y changing the laser mirrors. The control hassis, incorporating wide-band dc coupled ircuits for signal processing and blanking ind deflection circuits, was completed.

The SEM EBIC scan for the device without applied bias and the corresponding zero-bias video (unmodulated) 0.633 um response are shown in figure 20. The SEM was operated with an electron beam current of 2 nA and an accelerating voltage of 30 kV. It appears that most, if not all, of the features seen with the SEM can also be seen with the optical scanner. Presumably, the lines are striations in the silicon and the dark spots are swirls [37].

With the diode bias increased to yield a dark current of 1 or 2 μA, discrete spots of enhanced photoresponse to the unmodulated 0.633 um light could be seen at the junction periphery. The spots, which increased in brightness as the diode bias was further raised, are regions where the junction field at the surface is large enough to cause electrical breakdown with concomitant multiplication of the photoinduced charge carriers. However, observation of the 385- and 770-MHz 0.633 um photoresponse with the junction voltage raised to produce currents more than twenty times larger showed no such discrete photoresponse, or enhancement, at the periphery. Since ionic processes would not be expected to follow 385 MHz modulation, it is likely that the surface breakdown was ionic, rather than electronic, in nature.

When the detected modulating frequency was changed from 385 to 770 MHz, the optical photoresponse shrank and fell off more rapidly with distance from the connected bonding pad. This is because p-n junctions with contact areas less than junction areas lose their lumped R-C nature and increasingly become distributed networks as the signal frequency is raised. Applied to the present case, this has the consequence that the signal from charge carriers photogenerated and collected by the junction beyond the contact area is increasingly attenuated with distance from the contact. With the total junction capacitance known from independent low-frequency measurements, it should be possible to determine the sheet resistance, and perhaps local variations in the sheet resistance, of the p-diffused skin by interpreting the spatial and frequency dependence of the scanning photoresponse.

The response of the device to 1.15 um light is quite different from the 0.633 um response. Almost all of the response features were different, except of course for the shadows of the pads and lead wire. Although only the central diode of the array was electrically connected, the lead wire could be followed across the entire die by observing its shadow; even portions of the die not connected electrically yielded an apparent photoresponse. The shadows of the bonding pads on the other eight array diodes were also seen, and the die portions not covered by pads showed structure. It is likely that these phenomena are due to internal reflection within the silicon die and that the structure observed is due to irregularities in the die bond. Preliminary observations on a device with a known die bond void tend to confirm this hypothesis.

(D. E. Sawyer and D. W. Berning)

8.2. Automated Scanning Low Energy Electron Probe

The scanning low energy electron probe (SLEEP) [38] is an electron beam probing technique in which an electron beam is first accelerated (to provide beam definition) and then decelerated by a grid placed in front of the specimen to be probed. The SLEEP technique is inherently simple, involving a low energy (800 to 900 V) gun structure and

Naval Research Laboratory, Washington, D. C. 20375.

a standard vidicon electromagnetic beam focusing and deflection system. The specimen under investigation is scanned by the electron beam in the retarding field region. Only those electrons are collected whose energies are sufficient to overcome the local potential barrier at the specimen surface. This collected current is measured in the specimen-cathode circuit. Similarly, electrons with insufficient energy to overcome the local surface potential are reflected from the specimen and may be collected to form the mirror-mode operation. Thus, either the directly collected current or reflected current provides a surface potential map of the specimen. In the current program, which is intended to develop an automated SLEEP as a versatile diagnostic tool for determination of semiconductor resistivity, defect density, and oxide uniformity as well as a means to test complex integrated circuits, only the collectedcurrent mode is being investigated. A single crystal target, adjacent to the specimen under test, is used to provide absolute voltage measurements.

SLEEP can be used on-line for separate or combined measurements of wafer resistivity, wafer surface defect density, uniformity of the ratio of dielectric constant to thickness for oxide films, and for programmable "contactless" production oriented, functional testing and exercising of complicated LSI. Since the specimen is at, or near, ground potential, SLEEP provides significant advantages over a standard mirror microscope. In the latter use, the specimen must be at, or near, the cathode potential up to 20 kV below ground. Similarly, the ionization damage due to soft x-rays in high voltage systems is not a major problem with SLEEP. The SLEEP technique can also be used in an off-line mode in which it may be interfaced with other complementary diagnosti: techniques.

Initial efforts have concentrated on the design of the electron gun and acquisition of the computer control system. The design of the gun is similar to that of a commercial vidicon tube. The electrons from the cathode are accelerated by the grids and directed onto a cap at one end of the drift tube where the beam size is defined by an aperture 8 um in diameter. The specimen being probed is mounted 2 mm from the other end of the drift tube and the aperture is imaged on the surface of the specimen by magnetic focus and deflection as in a standard vidicon. (W. C. Jenkins* and G. P. Nelson