Table of contents

VLSI reliability issues such as hot-carrier effects, dielectrics and metallization, are reviewed and discussed. VLSIS have been developed mainly with continued reliability in mind. However, a new challenging approach to VLSI reliability is now greatly needed in response to the 'paradigm shift' now being brought about by simple scaling limitations, increased process complexity and VLSI application to advanced systems. A good example of this shift is the new movement from simple failure analysis by sampling the output of a manufacturing line to the 'building-in-reliability' approach. To pursue this technique, greater importance will be attached to a deeper physical understanding (including frequent use of CAD/DA) of the significant relationships between the input variables and product reliability, and to total concurrent engineering from research labs to production sites. Furthermore, fast new VLSI testing methods and new yield-enhancing redundancy techniques, resulting in cost reduction, will be increasingly needed to achieve high reliability for VLSIS with 10⁹ devices on a single chip. This review aims at understanding the physics underlying important reliability problems: hot-carrier effects in scaled MOSFETs, dielectrics in the insulator and electro/stress migration in the interconnection.

This article reviews reliability phenomena in thin silicon dioxide. We discuss a comprehensive framework for evaluating measured SiO₂ breakdown data which enables assurance of built-in oxide reliability for scaled MOS technologies. Promising technological improvements for improving SiO₂ reliability are also reviewed. We discuss an integrative view for explaining the many diverse observations about the process of oxide wear-out and failure.

The influence of surface technological processes is reviewed with special emphasis on increased device reliability and lifetime. Surface quality plays a significant role in many degradation mechanisms because it influences all technological processes such as etching, metal evaporation and annealing as well as surface passivation quality. It is shown that such technological hardening is extremely important and could perhaps be performed by suitable surface technological processes such as etching, cleaning and brief chemical passivation. Lifetimes are also limited by electrical stress during the fabrication of the devices, during assembly of systems and during operation. To get satisfactory device protection at microwave frequencies there is a need to protect circuits against electrostatic discharges and transients. The development and properties of a promising new concept based on field emission switches are presented.

The fundamentals of light-emitting phenomena from silicon semiconducting material are presented from an experimental rather than a theoretical point of view. The following aspects are considered in detail. In the first place, laterally resolved measurements give information about the distribution of emitted light, which has been generated by an excited radiative system. Secondly, spectrally resolved experiments enable a distinction to be made between different current transport mechanisms and permit discussion of underlying physical concepts. Thirdly, some fundamental emission mechanisms are discussed on the basis of the dependency of the electroluminescence intensity on such external parameters as temperature and current. Finally, the applicability of these mechanisms with respect to reliability physics, design verification and failure analysis of semiconducting devices is outlined and the state of the art of emission microscopy is reviewed.

The quasibinary compounds (Pb,Sr)Se and (Pb,Eu)Se are compared with each other and with PbSe regarding the shape of the interband optical absorption edge and the photoluminescence spectra. No differences in the edge are detected without magnetic field, if scaling with composition of the gap and of the effective masses is taken into account. Thus admixing of d states from the paramagnetic Eu²⁺ ions should not take place in the composition up to 10 at.% EuSe. In (Pb,Eu)Se an additional luminescence line was detected, and was attributed to transitions from the conduction band to the Eu²⁺ 4f states. In magnetic fields up to 7 T this line does not shift, whereas the interband line shifts with B. This shift cannot be explained by the Zeeman and spin splitting of the band states alone, but additional spin-spin interaction between the localized 4f and extended valence band states is proposed.

Hot electron to neutral acceptor (e, A⁰) photoluminescence from GaAs in strong magnetic fields up to 13 T is reported for the first time. The observed phonon cascade, and the changes in the emission arising from hot electrons recombining immediately after excitation, suggest that electrons relax by phonon-assisted scattering between bands dispersing freely in k_Z and stacked onto Landau levels. A small lineshift is observed, which can be explained quantitatively when non-parabolicities are taken into account, as is shown by separate measurements and with a k.p band structure calculation.

The band tail effect on the thermal emission (TE) integrated power and the spectral distribution ( lambda =3-5 mu m) in n-InAs (N_d-N_a=1.6*10¹⁶ cm^-3) has been investigated for the first time. The Urbach parameter T_eff has been determined in the temperature range 310-400 K. Some practical uses for the results obtained are also discussed.

This paper reports direct measurement of the total Cu concentration and its electrically active fraction in ultra-pure Ge crystals. The fraction of electrically active Cu atoms is shown to be 3*10^-3 in hydrogen-grown Ge and 3*10^-2 in vacuum-grown Ge. The state of Cu in ultra-pure Ge crystals and the passivation of electrically active Cu-related centres by hydrogen are discussed.

New high-resolution FTIR transmission spectra are presented for the fine structure of local vibrational modes associated with ¹⁰B, ¹¹B and ¹²C impurities in GaAs. Attempts are made at reproducing these results from cluster calculations based on a range of harmonic force constant models. For the boron impurities good agreement with the fine structure arising from host crystal isotope effects is readily achieved, but the impurity isotope shift cannot be satisfactorily reproduced by harmonic models which can be deemed physically reasonable. The same conclusion applies to the carbon results and, in this case, even the host isotope effects are not so well reproduced. We derive a general relation between impurity isotope shifts and the width of the fine structure from host crystal isotope effects. Both the detailed model calculations and this more general relationship point to the conclusion that, for any model to be satisfactory, anharmonicity must be included.

Thin films (t approximately 0.7 mu m thick) of ZnS_xCdSe_1-x (0<or=x<or=1) were formed by thermal evaporation on glass substrates kept at 300 K and 470 K. XRD studies showed that all the films were polycrystalline in nature. The wurtzite to zinc blends transition was found to take place in the composition region 0.70<or=x<or=0.75. The lattice parameters of the films formed at a higher substrate temperature were greater than those deposited at a lower substrate temperature. Optical bandgaps also increased with substrate temperature. Bandgaps varied nonlinearly with composition, showing a downward bowing. The bowing parameter decreased with substrate temperature. The structural and bandgap studies clearly indicated a complete miscibility of the ZnS-CdSe system.

A detailed study of electron traps in SIMOX buried oxides prepared by both single- and multiple-implant acid anneal technologies is reported. The techniques of avalanche electron injection and constant voltage stressing (CVS) have been used to reveal three distinct electron traps in SIMOX. The largest, with a capture cross section of the order 10^-14 cm², is thought to be related to the so-called E' centres associated with excess Si in the oxide matrix. Its concentration is seen consistently to be higher in single-implant material which is known to contain a significantly higher concentration of excess Si than multiple-implant material. SIMOX also contains a smaller trap, with a cross section typically of the order 10^-16 cm², and it is suggested that it is a result of contamination during implantation. The third and smallest trap has a cross section of the order 10^-17 cm² and may be the so-called water-related trap seen in thermal oxides. The experiments also show that useful quantitative information can be obtained from the relatively crude but simple technique of constant voltage stressing. This is mainly because the field necessary to induce significant electron current through the oxide is significantly lower than in conventional thermal oxide as the injection is aided by field intensification arising from interfacial asperities. CVS has therefore been developed into a simple oxide assessment technique which could be readily applied to starting substrates on simple, mesa-defined MOS capacitors.

Strong signals corresponding to electron paramagnetic resonance (EPR) transitions with the absorption of two, three and so on microwave quanta have been observed in optically detected magnetic resonance (ODMR) spectra of bulk semiconductors (A²B⁶, SiC), type-II GaAs/AlAs superlattices and silver chloride crystals. Multiquantum ODMR spectroscopy was applied to analyse the energy levels of Al, Ga and Sc accepters in silicon carbide and showed a nonlinearity of the energy levels of Sc acceptors in a magnetic field. In ASCI crystals, multiquantum ODMR of self-trapped holes, electrons and self-trapped excitons was found. For the first time, multiquantum ODMR was observed between fine-structure energy levels, i.e. for a system with electron spin S>1/2. Singlet-to-triplet multiquantum EPR transitions with absorption of up to seven microwave quanta were detected for self-trapped excitons and unambiguously identified by frequency shift measurements. The singlet-to-triplet splitting has been determined with a very high accuracy: J=-161.0+or-0.1 GHz.

The development of structural defects in In_xGa_1-xAs/GaAs strained-layer structures, grown by molecular beam epitaxy, was investigated by transmission electron microscopy. Three different critical layer thicknesses, corresponding to the formation of three different types of defects, were observed with increasing thicknesses of the strained layers. These defects were: rough interfacial topographies resulting from an onset of three-dimensional growth for the In_xGa_1-x layers; misfit dislocations of the 60 degrees mixed type; and dislocation complexes consisting of planar defects on (111) planes.

Low-temperature time-resolved degenerate four-wave mixing (DFWM), single-beam excite/probe and transient grating measurements have been performed on In_xGa_1-xAs/GaAs superlattices with thin ( approximately 6 nm) In_xGa_1-xAs wells. We find weak dephasing of heavy-hole (hh) excitons by acoustic phonons. The hh exciton homogeneous linewidth temperature coefficient is 1.9+or-1 mu eV K^-1. The low-temperature hh exciton optical nonlinearity is successfully modelled in terms of exciton phase-space filling and exchange interactions in the hh exciton gas. From low-temperature transient grating measurements we find hh exciton in-plane mobilities as high as 6.7*10⁴ cm² V^-1 s^-1 and mobilities which are greater in superlattices with thinner barriers.

The effect of the space correlation of oxide charges on the effective mobility of electrons in the channel of an NMOS transistor has been studied. Mobility has been obtained by a single-electron Monte Carlo simulation in which space correlation has been included simultaneously with other effects such as screening of point charges by mobile carriers, image charges, and phonon and surface-roughness scattering. Results have been obtained by assuming different degrees of space correlation for several temperature values and diverse oxide-charge concentrations and positions. The charged-centre space correlation is shown to have a noticeable effect on the electron mobility.

Exact analytical expressions for the switching delay of an inverter driving an RC load, taking into account the velocity saturation, are obtained. Modified expressions to include the effect of source resistance are then presented. Owing to the limitation on switching current imposed by the velocity saturation mechanism, the switching delay is substantially increased for identical width-to-length ratios of the MOSFET in a complementary MOS logic circuit. Obviously, it is important to increase the saturation velocity by miniband engineering or otherwise to improve the performance. However, it is found that the improvements in the time delay are marginal after a saturation velocity of 2.4*10⁷ cm s^-1 is reached. The effect of technologies aimed at circumventing the hot-electron and other deleterious effects is a longer delay time of the circuit due to increased series resistance. However, the effect of increased series resistance is substantially damped due to velocity saturation. These results are particularly important in designing CMOS circuits with submicrometre MOSFET dimensions.

A new unified device model of a polysilicon emitter transistor (PET) is proposed for various emitter structures. More factors influencing the base current are considered in this model and the general expressions for the common emitter current gain (h_FE) and the emitter Gummel number (G_e) are derived in an analytical form. From these general expressions we can classify five different PET device structures of PET by specifying some structure parameters of the devices. Ning's theoretical equation for the current gain enhancement factor K and De Graaff's theoretical equation for G_e can be deduced directly from this model. This model is in good agreement with experimental results. The value for the minority hole mobility in polysilicon is also deduced to be in the range of 4.5-6.1 cm² V^-1 s^-1 from fitting to the experimental data.

The three-phase-shift (3PS) distributed feedback (DFB) laser is known to have improved performance over the lambda /2 DFB laser because the spatial hole burning effect is suppressed. When optimizing the structural design of the SPS DFB laser, the threshold gain difference and the field distribution along the laser cavity have been selected as the design criteria. Using the transfer matrix method, it is found that the optimum phase-shift values and their positions are in trade-off when one attempts to optimize both the gain difference and field uniformity simultaneously.

Photovoltaic in detectors have been fabricated by a photolithographic process on lead selenide (PbSe) for near ambient temperature operation (200-300 K) in the 3-5 mu m spectral band. The PbSe active layer has been grown by molecular beam epitaxy directly on a Si(111) substrate using a CaF₂ buffer layer. The bulk, lateral and surface electrical effects which affect sensor performances have been investigated on variable-area diodes. The dark current is dominated by generation-recombination surface currents which are probably due to an inefficient surface passivation. The lateral collection of photocarriers increases the efficiency of the sensor: at 300 K, the collection length is around 70 mu m and the derived specific detectivity is around 3*10⁹ cm Hz^1/2 W^-1.

A series of GaSb single crystals doped with manganese was grown by the Czochralski method without encapsulant in a flowing atmosphere of hydrogen. From Hall and conductance measurements it has been found that manganese creates shallow acceptors with an activation energy of 16+or-2 meV, which strongly enhance conductivity at low temperatures. Using the values of Hall concentrations in the samples cut from the grown crystals, the manganese distribution coefficient in GaSb was calculated to be 0.011+or-0.001 for Mn concentration in the melt within the range 7.0*10²⁴-9.7*10²⁵ atoms/m³.

During non-selective wet chemical gate recess etching of pseudomorphic In_0.15GaAs/GaAs HEMTs, the rate of change of saturated drain current with etch time decreased suddenly to one-tenth of the initial rate when etching began. This reduced rate lasted for around 45 s of etching before the rate increased again. A simple model, based on Poisson's equation with a constant etch rate in all the layers of the HEMT structure, has been developed to explain the effect. This shows that the variation in the rate of change of drain current with etch time is related to the thickness and doping concentration of the layers in the HEMT structure. As will be explained, this has important ramifications for HEMT layer design.

We report the application of the reflectance difference (RD) technique as an in situ probe to study the wet etching process of GaAs. RD responses for different etch solutions and substrate types are studied. The signals are most pronounced for acidic hydrogen peroxide-containing etch solutions. The existence of strong anisotropies on the etched surface aligned with crystallographic orientations at the initial stage indicates that the etching process starts with the creation of a well defined outermost top layer of only one type of bulk atom. A comparison between semi-insulating and n-type substrates shows that the formation of anisotropies during the etch process is enhanced for n-type substrates. Our experiments show that the RD technique is suitable for in situ investigation of surface processes even in liquid environments and gives new information about the wet etching process.

An analysis of the effects of a magnetic field applied perpendicular and parallel to the initial injection velocity of ballistically injected energetic electrons in a quasiballistic heterojunction bipolar transistor has resulted in an expected reduction in electron energy directed towards the collector barrier. A measure of the high energy effective mass in InGaAs resulted in a value of 0.08 times the free electron mass. In addition, the application of a magnetic field parallel to the ballistic velocity induces a broadening and shift to higher energies of the thermalized distribution. This result has been interpreted in terms of a reduction in the scattering rate of energetic electrons due to the imposition of quasi one-dimensional constraints by the magnetic field.