Philips (Belgium)

Numerical methods are described for determining robust, or well-conditioned, solutions to the problem of pole assignment by state feedback. The solutions obtained are such that the sensitivity of the assigned poles to perturbations in the system and gain matrices is minimized. It is shown that for these solutions, upper bounds on the norm of the feedback matrix and on the transient response are also minimized and a lower bound on the stability margin is maximized. A measure is derived which indicates the optimal conditioning that may be expected for a particular system with a given set of closed-loop poles, and hence the suitability of the given poles for assignment.

During the last half century, a dramatic downscaling of electronics has taken place, a miniaturization that the industry expects to continue for at least a decade. We present efforts to use the self-assembly of one-dimensional semiconductor nanowires1 in order to bring new, high-performance nanowire devices as an add-on to mainstream Si technology. The nanowire approach offers a coaxial gate-dielectric-channel geometry that is ideal for further downscaling and electrostatic control, as well as heterostructure-based devices on Si wafers.

The algebraic theory of linear time-invariant systems has been studied in large detail during the past few decades and numerous computational algorithms have been developed to solve problems arising in this context. In this paper the numerical aspects of a certain class of such algorithms-dealing with what the author calls generalized eigenstructure problems-are discussed. Some new and/or modified algorithms are presented. Both the nmnerical stability of the algorithms and the conditioning of the problems they solve are analyzed using numerical criteria.

A fast algorithm is presented for deciphering cryptograms involved in the public-key cryptosystem proposed by Rivest, Shamir and Adleman. The deciphering method is based on the Chinese remainder theorem and on improved modular multiplication algorithms.

The RF noise in 0.18-/spl mu/m CMOS technology has been measured and modeled. In contrast to some other groups, we find only a moderate enhancement of the drain current noise for short-channel MOSFETs. The gate current noise on the other hand is more significantly enhanced, which is explained by the effects of the gate resistance. The experimental results are modeled with a nonquasi-static RF model, based on channel segmentation, which is capable of predicting both drain and gate current noise accurately. Experimental evidence is shown for two additional noise mechanisms: 1) avalanche noise associated with the avalanche current from drain to bulk and 2) shot noise in the direct-tunneling gate leakage current. Additionally, we show low-frequency noise measurements, which strongly point toward an explanation of the 1/f noise based on carrier trapping, not only in n-channel MOSFETs, but also in p-channel MOSFETs.

The statistical use of a particular classic form of a connectionist system, the multilayer perceptron (MLP), is described in the context of the recognition of continuous speech. A discriminant hidden Markov model (HMM) is defined, and it is shown how a particular MLP with contextual and extra feedback input units can be considered as a general form of such a Markov model. A link between these discriminant HMMs, trained along the Viterbi algorithm, and any other approach based on least mean square minimization of an error function (LMSE) is established. It is shown theoretically and experimentally that the outputs of the MLP (when trained along the LMSE or the entropy criterion) approximate the probability distribution over output classes conditioned on the input, i.e. the maximum a posteriori probabilities. Results of a series of speech recognition experiments are reported. The possibility of embedding MLP into HMM is described. Relations with other recurrent networks are also explained.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

A solution is proposed to the long-standing problem of the numerical instability of fast recursive least squares transversal filter (FTF) algorithms with exponential weighting, an important class of algorithms for adaptive filtering. A framework for the analysis of the error propagation in FTF algorithms is first developed; within this framework, it is shown that the computationally most efficient 7N form is exponentially unstable. However, by introducing redundancy into this algorithm, feedback of numerical errors becomes possible; a judicious choice of the feedback gains then leads to a numerically stable FTF algorithm with a complexity of 8N multiplications and additions per time recursion. The results are presented for the complex multichannel joint-process filtering problem.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

The classical Levinson algorithm for computing the predictor polynomial relative to a real positive definite Toeplitz matrix is shown to be redundant in complexity. It can be broken down into two simpler algorithms, either of which needs only to be processed. This result can be interpreted in the framework of the theory of orthogonal polynomials on the real line as follows: the symmetric and antisymmetric parts of the predictors relative to the sequence of Toeplitz matrices constitute two families of polynomials orthogonal on the interval [- 1,1] with respect to some even weight functions. It turns out that the recurrence relations for these orthogonal polynomials can be used efficiently to compute the desired predictor. The resulting "split Levinson algorithm" requires roughly one-half the number of multiplications and the same number of additions as the classical Levinson algorithm. A simple derivation of Cybenko's method for computing the Pisarenko frequencies is obtained from the recurrence relations underlying the split Levinson algorithm.

On annealing a boron implanted Si sample at ∼800 °C, boron in the tail of the implanted profile diffuses very fast, faster than the normal thermal diffusion by a factor 100 or more. After annealing for a sufficiently long time, the enhanced diffusion saturates. The enhanced diffusion is temporary, on annealing the sample a second time after saturation, enhanced diffusion does not occur. It is therefore designated as transient enhanced diffusion (TED). The high concentration peak of the implanted boron profile, which is electrically inactive, does not diffuse. TED makes it difficult to fabricate modern Si based devices, in particular TED produces the parasitic barriers which degrade the performance of the SiGe heterostructure bipolar transistors and TED can limit the fabrication of shallow junctions required for sub-100 nm complementary metal–oxide–semiconductor technology. The mechanisms of TED have been elucidated recently. A Si interstitial “kicks out” the substitutional boron atom to an interstitial position where it can diffuse easily. Alternatively the interstitials and boron atoms form highly mobile pairs. In both cases Si interstitials are required for the diffusion of boron. Therefore the enhanced boron diffusivity is proportional to the concentration of the excess Si interstitials. The interstitials are injected during implantation with Si or dopant ions. The interstitials are also injected during oxidation of the Si surface. Therefore the diffusivity increases temporarily in both cases. Even at relatively low annealing temperatures (∼800 °C) the mobility of the interstitials is high. The TED at this temperature lasts for more than 1 h. This large TED time can be explained by the presence of interstitial clusters and interstitial–boron clusters. The interstitial clusters are the {311} extended defects and dislocation loops. The precise structure of interstitial–boron clusters is not yet known though several models have been proposed. The clusters are the reservoirs of the interstitials. When the supersaturation of interstitials becomes low, the clusters dissolve and emit interstitials. The interstitials emitted from the clusters sustain the TED. Many groups have suggested that the rate of emission of interstitials is determined by Ostwald ripening of the clusters. However, recently TED evolution has also been explained without invoking Ostwald ripening of the {311} defects. The evidence of Ostwald ripening of dislocation loops is more direct. In this case the Ostwald ripening has been confirmed by the measurements of the size distributions of the dislocation loops at different times and temperatures of annealing. At higher temperatures the extended clusters are not stable and coupling between the interstitials and boron atoms is reduced. Therefore at high temperatures TED lasts only for a short time. At high temperatures the displacement during TED is also small. This suggests that if rapid thermal annealing with high ramp rates is used, TED should be suppressed. Currently high ramp rates, 300–400 °C/s are being tried to suppress TED.

We have done an empirical study of the relation of the number of parameters (weights) in a feedforward net to generalization performance. Two experiments are reported. In one, we use simulated data sets with well-controlled parameters, such as the signal-to-noise ratio of continuous-valued data. In the second, we train the network on vector-quantized mel cepstra from real speech samples. In each case, we use back-propagation to train the feedforward net to discriminate in a multiple class pattern classification problem. We report the results of these studies, and show the application of cross-validation techniques to prevent overfitting.

A theoretical analysis is made of the error propagation due to numerical roundoff for four different Kalman filter implementations: the conventional Kalman filter, the square root covariance filter, the square root information filter, and the Chandrasekhar square root filter. An experimental analysis is performed to validate the new insights gained by the theoretical analysis.

Relates three classical concepts, viz. mixed radix number system, Kronecker product of matrices, and perfect shuffle. It presents an algebra which describes the hardware organization of the computation of a product <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">M</i> <b xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">v</b> , where <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">M</i> is a matrix in Kronecker product form and <b xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">v</b> is a vector. The algebraic formalism describes both the blockwise structure of the computation and the various possible connection patterns.

Abstract The paper is concerned with applications of the standard Nevanlinna–Pick problem in various technical domains, namely the following: interpolation by reflectance functions, polynomial stability checking, cascade synthesis of passive one‐ports, and model reduction with a Hankel norm criterion. Some fundamental results on the Nevanlinna–Pick problem are shown to be of a definite interest in each of these subjects.

A phoneme based, speaker-dependent continuous-speech recognition system embedding a multilayer perceptron (MLP) (i.e. a feedforward artificial neural network) into a hidden Markov model (HMM) approach is described. Contextual information from a sliding window on the input frames is used to improve frame or phoneme classification performance over the corresponding performance for simple maximum-likelihood probabilities, or even maximum a posteriori (MAP) probabilities which are estimated without the benefit of context. Performance for a simple discrete density HMM system appears to be somewhat better when MLP methods are used to estimate the probabilities.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Models of large and complex systems can often be reduced to smaller sub-models, for easier analysis, by a process known as decomposition. Certain criteria for successful decompositions can be established.

Comparison of digital and analog figures-of-merit of FinFETs and planar bulk MOSFETs reveals an interesting tradeoff in the analog/RF design space. It is found that FinFETs possess the following key advantages over bulk MOSFETs: reduced leakage, excellent subthreshold slope, and better voltage gain without degradation of noise or linearity. This makes them attractive for digital and low-frequency RF applications around 5 GHz, where the performance-power tradeoff is important. On the other hand, in high-frequency applications, planar bulk MOSFETs are seen to hold the advantage over FinFETs due to their higher peak transconductance. However, this comes at a cost of a reduced voltage gain of bulk MOSFETs

Principles, Problems and Approaches. The Deterministic Approach. The Statistical Approach. Thermodynamic Extension. Higher Order Networks. Network Design. Bibliography. Index.

This paper presents reconfigurable RF integrated circuits (ICs) for a compact implementation of an intelligent RF front-end for multiband and multistandard applications. Reconfigurability has been addressed at each level starting from the basic elements to the RF blocks and the overall front-end architecture. An active resistor tunable from 400 to 1600 /spl Omega/ up to 10 GHz has been designed and an equivalent model has been extracted. A fully tunable active inductor using a tunable feedback resistor has been proposed that provides inductances between 0.1-15 nH with Q>50 in the C-band. To demonstrate reconfigurability at the block level, voltage-controlled oscillators with very wide tuning ranges have been implemented in the C-band using the proposed active inductor, as well as using a switched-spiral resonator with capacitive tuning. The ICs have been implemented using 0.18-/spl mu/m Si-CMOS and 0.18-/spl mu/m SiGe-BiCMOS technologies.

The shot noise, line edge roughness (LER) and quantum efficiency of EUV interaction with seven resists related to EUV-2D (SP98248B) are studied. These resists were identical to EUV-2D except were prepared with seven levels of added base while keeping all other resist variables constant. These seven resists were patterned with EUV lithography, and LER was measured on 100-200 nm dense lines. Similarly, the resists were also imaged using DUV lithography and LER was determined for 300-500 nm dense lines. LER results for both wavelengths were plotted against E_size. Both curves show very similar LER behavior-the resists requiring low doses have poor LER, whereas the resists requiring high doses have good LER. One possible explanation for the observed LER response is that the added base improves LER by reacting with the photogenerated acid to control the lateral spread of acid, leading to better chemical contrast at the line edge. An alternative explanation to the observed relationship between LER and E_size is that shot-noise generated LER decreases as the number of photons absorbed at the line edge increases. We present an analytical model for the influence of shot noise based on Poisson statistics that preidicts that the LER is proportional to (E_size)^-1/2. Indeed, both sets of data give straight lines when plotted this way (DUV r² = 0.94; EUV r² = 0.97). We decided to further evaluate this interpretation by constructing a simulation model for shot noise resulting from exposure and acid diffusion at the mask edge. In order to acquire the data for this model, we used the base titration method developed by Szmanda et al. to determine C-parameters and hence the quantum efficiency for producing photogenerated acid. This information, together with film absorptivity, allows the calculation of number and location of acid molecules generated at the mask edgte by assuming a stochastic distribution of individual photons corresponding to the aerial image function. The edge "roughness" of the acid molecule distribution in the film at the mask edge is then simulated as a function of acid diffusion length and compared to the experimental data. In addition, comparisoins between of the number of acid molecules generated and photons consumed leads to values of quantum efficiencies for these EUV resists.

The Hidden Markov models are generalized by defining a new emission probability which takes the correlation between successive feature vectors into account. Estimation formulas for the iterative learning both along Viterbi and Maximum likelihood criteria are presented.