N. Nintunze

Programmable impulse neural circuits

A description is given of CMOS electronic circuits which emulate natural neurons at a more detailed level than that typically used by artificial neural network models. A pulse-firing circuit which realizes general short-term neuron dynamics is discussed. Both fixed and programmable synapse circuits for realizing long-term dynamics are also described. Together, these establish the basic structures required for the implementation of programmable impulse neural networks.

Hole drift velocity in the warped band model of GaAs

Hole transport in the warped band model of GaAs has been investigated using the ensemble Monte Carlo approach. Scattering rates used in the simulations were derived for warped heavy and light hole bands and for a spherical split-off band. Warping was taken into account exactly in the determination of carrier states after scattering. The calculated velocity-field characteristics are in better agreement with experimental data than models that account for valence band warping by the use of approximate overlap functions. The valence band parameter set (A=-7.65, B=-4.82, C=7.70) gave the closest fit to the experiment and to results of a realistic band-structure model.

ULTRAFAST ENERGY LOSS OF ELECTRONS IN P-GAAS

We investigated the ultrafast energy relaxation of photoexcited minority electrons in highly doped p‐GaAs by means of femtosecond time resolved luminescence (Δt<90 fs). Our experiments allow the first observation of the extremely fast cooling of minority electrons within the Γ‐valley. The electron mean energy decreases within 150 fs from 150 meV down to less than 50 meV. The total energy loss rate reaches values higher than 10−7 W per electron, representing the highest energy loss rates of electrons observed to date in monocrystalline semiconductors. Ensemble Monte Carlo simulations show that the electron‐hole scattering is responsible for these high energy loss rates.

VLSI implementation of a pulse Hebbian learning law

The authors describe a Hebbian learning rule which is particularly well suited for VLSI implementation. The authors first introduce a signal coding method where signal information is represented as pulse probabilities. The authors then briefly describe the functionality of an impulse neuron circuit in terms of that method. Competitive learning is then reviewed in the context of the probabilistic representation. The pulse Hebbian learning law is then introduced and analyzed in these terms. The authors close with an example of a simple pulse Hebbian synapse circuit for use in a competitive neural network.<<ETX>>

A wide-dynamic-range programmable synapse for impulse neural networks

A fully-analog synapse circuit for impulse neural networks is described. The circuit is programmable over a dynamic range of several orders of magnitude. It is essentially a programmable switched-circuit source which incrementally contributes to the excitation of an impulse neuron. Shared-floating-gate FETs are used to implement a programmable current mirror exhibiting a nonvolatile storage characteristic. The synapse connection strength is modulated by incrementally adding or removing charge on the shared floating gate. A wide dynamic range is achieved through biasing the circuit to operate in the subthreshold region.<<ETX>>

Carrier-carrier interaction effects on transient valley repopulation and velocity in silicon

The effects of carrier-carrier interactions on the transient valley repopulation and velocity in bulk silicon has been investigated using an ensemble Monte Carlo method. The expressions for electron-electron scattering rates take into account the ellipsoidal nature of the energy surfaces in silicon. The relative effects of both electron-plasmon and short range e-e interactions strongly depend on the orientation and magnitude of the electric fields. The e-e interactions reduce the transfer rates of electrons from the hot valleys to the cold valleys, increase the energy of the electrons in the cold valleys, and reduce the transient velocity overshoot in the hot valleys. These effects are more significant at higher electron concentration.

Diffusion Coefficient of Electrons in Diamond

Abstract The diffusion coefficient of electrons in diamond has been investigated using Ensemble Monte Carlo procedure as a function of field strength and temperature. The electric field is assumed to be along the crystalographic orientation. It is found that the diffusion coefficient decreases at high fields when the nonparabolicity of the conduction band is taken into account and also decreases with temperature. The transverse diffusion coefficient is smaller than the longitudinal one at lower fields and is higher beyond 4 kV/cm.

Sub-Picosecond Luminescence Spectra of Photoexcited Electrons Relaxation in p-GaAs

Femtosecond lasers have allowed closer examination of the dynamics of photoexcited carriers in semiconductors during the first tens of femtosecond after laser excitation1–3. In particular, the ultrafast thermalization and relaxation of minority photoexcited carriers in p-GaAs have been a subject of intensive experimental investigations. For example, Furuta and Yoshii2 used time resolved photoluminescence to examine minority electron relaxation over a wide range of acceptor concentrations and demonstrated the importance of electron-hole interaction in the relaxation of the photoexcited minority electrons at high acceptor concentrations. However, because the energy used in the photoexcitation excitation process (1.9 eV) generated electrons with energies above the intervalley transfer threshold, energy loss through intervalley phonon emission contributed to the overall energy and momentum relaxation process. More recently, Rodrigus and coworkers3 also examined the relaxation of photoexcited minority electrons in p-GaAs using shorter laser pulses with energies below the intervalley transfer process in order to minimize the role of intervalley phonons and examine the role of e-h in the cooling process. They also compared the measured electron temperatures with Monte Carlo simulations and obtained good agreement between the measured and simulated data.

Femtosecond relaxation of highly photo-excited carriers in GaAs

Ultrafast relaxation of photo-excited electrons in p-doped and intrinsic GaAs has been investigated using the Monte Carlo method. Dynamic screening of the carrier-carrier (c-c) interaction has been implemented using a momentum and frequency dependent dielectric function. Compared to the static c-c scattering model, the current approach results in faster cooling of the electron-hole plasma, due to enhanced carrier-carrier scattering rates. In p- GaAs, the energy relaxation shows that the electron-hole plasma (EHP) cools faster with increasing hole concentration. The transient luminescence intensity and the effective carrier temperature computed from luminescence spectra compare favorably with experimental data.

Momentum reorientation of photo-excited carriers in GaAs

Momentum relaxation of photo-excited carriers in GaAs was investigated using the Monte Carlo approach. A laser of 1.51 eV photon energy and 9 fs width was assumed. Simulations were performed for excitation densities ranging from 1016 cm-3 to 8 X 1017 cm-3. For nexc equals 8 X 1017 cm-3, the distribution of the carrier momentum was found to approximate a Maxwell-Boltzmann distribution 25 fs after laser excitation, which concurs with recent experimental data. The relaxation time was shown to increase with decreasing carrier density and to be shorter when the carrier-carrier scattering was treated dynamically rather than statically.