Search

Silicon interposer (Si-IP) technology has been used in accelerated processing units such as graphic processing units in high-performance computing because it can package a system-on-chip and high bandwidth memories. However, the conventional Si-IP has difficulty developing larger packages because of the mismatch in the coefficient thermal expansions (CTE) of the Si-IP and the organic substrate. Therefore, the Si-IP has limited capacity for improving computing performance by the application which requires more chips. We developed a multilayer glass substrate (Glass-ST) that features a stacked glass core and propose to apply this Glass-ST to a computer board. The proposed structure has no CTE mismatch and can use high density wiring. Thus, the Glass-ST enables the assembly of more large chips than is possible using the conventional Si-IP. In this study, we prepared a 100X100 mm Glass-ST with a 5/5 µm line/space and 20 µmΦ vias. We mounted nine 21X21 mm chips with 40 µm pitch micro bumps. The results revealed that conformal plated through glass vias and a fine wiring pattern had been fabricated in the Glass-ST, and that the nine chips and Glass-ST were connected by micro bumps. The maximum warpage of the nine chips was 23 µm between temperatures of 30°C and 250°C. This means that the Glass-ST can mount chips with micro bumps due to the very slight resulting warpage. In addition, we performed thermomechanical simulation to investigate the stress experienced by the micro bumps. The results show that the maximum stresses of micro bumps with pitches ranging between 10 µm and 55 um are very similar to that of 40 µm pitch micro bumps with which the real sample was packaged. We believe the improvements in the computing performance are significant by the Glass-ST technology compared to that of the conventional Si-IP technology.

Embedding thin-film capacitors (TFC) in a package substrate is a technology aimed at improving the performance of power supply. The package substrate that we have developed and produced embeds TFCs into the core layer of a built-up substrate. Using this substrate promises to dramatically reduce the impedance on power supply lines and provide a stable power supply to the integrated circuit (IC) chip, resulting in improved performance of electronic devices. In this paper, we propose three improved structures for package substrates with embedded TFCs: the first is an asymmetric single-sided embedded TFC structure that uses the warpage-control layer; the second is a standalone embedded TFC structure that places solitary TFCs directly under the IC chip only; the third is a coreless embedded TFC structure that embeds TFCs in a coreless substrate. These technologies promise to help lower the cost of package substrates, increase their efficiency, and reduce their thickness.

High-density packaging technologies such as 2.5D interposer technology have the potential to enhance computing performance because interposer technologies can improve input/output connections for multi-chip packaging. However, it is difficult to use interposer technology to manufacture large-sized packages because of coefficient of thermal expansion (CTE) mismatches between the interposer and organic package substrate. Here, we reported a novel glass package substrate that is stacked with glass cores, and we used conductive paste with interstitial vias. The glass package substrate enables eliminating the packaging process of the interposer and low CTE of glass; thus, it is cost-effective for the printed circuit board process. The glass package substrate passed the reflow test five times, an unbiased highly accelerated stress test 300 hours, and the thermal cycle test 750 cycles. The warpage of the glass package substrate was smaller than that of conventional organic packages. The CTE of glass was the same as that of silicon; therefore, the glass package substrate can eliminate CTE mismatch. Moreover, the stress resistance for bending stress test was stronger because of the structure of the glass core. We believe that glass package substrates are suitable alternatives to interposer technologies.

2.5D and 2.1D technologies aim at improving wiring density and packaging density. However, both 2.5D and 2.1D technologies present coefficient of thermal expansion (CTE) mismatch issues and an increase in the assembly process. To overcome such limitations, we developed a glass package stacked glass core which uses the conductive paste as interlayer connection vias. We fabricated and then studied a multilayer glass package substrate. The daisy chain of such multilayer glass package substrate passed through a reflow test for five times. During the reflow process, the warpage of the multilayer glass package substrate turns to be lower than that of an organic package substrate. In addition, the CTE of the multilayer glass package substrate is approximately 3 ppm/C, as in silicon. The bending strain of the multilayer glass package substrate becomes stronger as the number of layers increase and its internal stress is lower than the glass interposer. We believe that the proposed multilayer glass package substrate will guarantee high-density wiring, with high reliability and high packaging density, and that it can be optimally used in practical applications instead of 2.5D and 2.1D technology.

This paper reports on a large-size CPU package for UNIX servers which employs embedded thin film capacitor layers. The substrate of this package has two thin film capacitor layers in the surface of the core layer, which has a capacitance of 25 uF in total. In order to adopt this package substrate, we confirmed the effect of the thin film capacitor layers on the package assembly process. We actually measured mechanical characteristics and the coefficient of thermal expansion (CTE), and confirmed that we can use it like conventional package substrates. This package has passed the preprocessing and subsequent environmental test on JEDEC Level 4, and we think that a high-quality package has been developed this time.

This paper reports on a large-size CPU package for UNIX servers which employs embedded thin film capacitor layers. The substrate of this package has two thin film capacitor layers in the surface of the core layer, which has a capacitance of 25 uF in total. In order to adopt this package substrate, we confirmed the effect of the thin film capacitor layers on the package assembly process. We actually measured mechanical characteristics and the coefficient of thermal expansion (CTE), and confirmed that we can use it like conventional package substrates. This package has passed the preprocessing and subsequent environmental test on JEDEC Level 4, and we think that a high-quality package has been developed this time. Fujitsu adopted a low melting point solder BGA from a previous-generation package. Its composition is Sn-57Bi-1.0Ag. However, this solder has an issue with deformation after a low-temperature reflow profile for the solder ball attachment process on the package. In this work, we were able to find the cause of this phenomenon, and by changing the Ag content from 1.0% to 0.4%, we achieved a complete low-temperature assembly process. In addition, BGA connection reliability was confirmed.

This paper proposes a method of improving power delivery in a CPU package substrate with multilayer thin film capacitors (TFCs). Because TFC embedded technology can reduce the inductance of power delivery path, it can stabilize the power supply to the CPU. Higher permittivity, thinner layer thickness and wider conductor plane area results in larger capacitance. To further stabilize the power supply, we developed a package structure with an embedded double TFC layer that can double the capacitance. The double TFC embedded package was formed by laminating two TFCs to the top and bottom of the core layers in a build–up substrate. Through verification using test vehicles, it was confirmed that electrical performance was significantly improved with the double TFC. Also, it was demonstrated that the criteria of the capacitance and insulation resistance were satisfied in the product reliability tests.

For energy harvesting systems, a novel charge pump with cross-connected structure is proposed in this paper. Owing to the cross-connected structure, the proposed charge pump can offer the output voltage to the output load at every phase. Furthermore, the proposed charge pump can reduce the number of circuit stages from the conventional charge pump. For above-mentioned reasons, the proposed charge pump can realize not only smaller internal resistance but also smaller output capacitance than the conventional charge pump. The theoretical analysis and simulation program with integrated circuit emphasis (SPICE) simulation demonstrate that the proposed charge pump outperforms the conventional charge pump in the point of power efficiency and circuit speed.

A new organic build-up substrate packaging technology was developed in Fujitsu for high-end servers, where lower V-G impedance on substrate and thermal resistance are realized by applying metallic thermal injection materials. In the present paper, two important accomplishments in assembling process, eliminating voids with metallic thermal injection and controlling substrate flatness on mounting large LSIs, are investigated, and evaluated.

Cantonese is a well-known Chinese dialect with a quite complex tone system. We have successfully applied the command-response model to represent F/sub 0/ contours of Cantonese speech by defining a set of appropriate tone command patterns. It provides an efficient means to describe Cantonese F/sub 0/ contours with high accuracy. Both qualitative and quantitative descriptions of the command patterns of nine tones are presented. The various cues for identifying each tone are investigated, based on which a set of rules is derived to synthesize F/sub 0/ contours of Cantonese. Perceptual experiments have been conducted to test the validity of the rules for each tone and to evaluate the naturalness of synthetic speech based on those rules.

Cantonese is well-known as a Chinese dialect with a highly complex tone system. Like Mandarin, the tones in Cantonese are traditionally denoted by a 5-scale notation system, which cannot provide a quantitative model for the F/sub 0/ contours of continuous speech. This paper applies the command-response model, which has already been shown to apply quite well to Mandarin, to represent the F/sub 0/ contours of Cantonese. A set of appropriate tone command patterns is derived from analysis-by-synthesis. By intrinsically incorporating the effect of tone coarticulation word accentuation and phrase intonation, the model can provide a high accuracy of approximation to the F/sub 0/, contours of Cantonese, and hence can serve as a much better tool than the traditional 5-scale tone notation. The success of the model for both Mandarin and Cantonese suggests that the model can be efficiently applied to various tone languages.

Cantonese is a well-known Chinese dialect with a quite complex tone system. We have applied the command-response model to represent F/sub 0/ contours of Cantonese speech by defining a set of appropriate tone command patterns. In this paper, the analysis is extended to Cantonese utterances at three different speech rates. By incorporating the effects of tone coarticulation, word accentuation and phrase intonation, the model gives high accuracy of approximations to Cantonese speech F/sub 0/ contours, and hence provides a much better means to quantitatively describe the F/sub 0/ contours than the traditional 5-level tone code system. The distributions of timing and amplitudes of commands are investigated, based on which a set of rules is used for synthesis of Cantonese F/sub 0/ contours. The validity of the current approach is confirmed by perceptual evaluation of Cantonese synthetic speech.

The Linear Predictive (LP) method has been widely used in speech analysis, mainly because of the simple mathematical formulation of the model and the straightforward computation of its parameters. However, there still remain certain difficulties that cause errors in the result of the analysis. Often encountered are the errors due to harmonic structure of the excitation source. The Spectral Transform LP (STLP) method proposed in the present paper aims at reducing these errors. Amplitude transforms on the input spectrum and on the spectrum of the model are introduced to modify the error criterion and the model adopted in the standard LP analysis. We show by analyses of both synthetic and natural speech that the STLP method offers significant improvement over the standard LP method. A method of STLP speech synthesis using the standard LP model is proposed. A perceptual experiment confirms the superiority of the STLP method in analysis of speech.

Critical re-examination of the premises of conventional systems for continuous speech recognition has led to a study of human processes of speech perception. It was found that deletion of a syllable is often not noticed by a listenter, suggesting that the basic unit of continuous speech perception is larger than the syllable. Further experiments are described and discussed on the size of actual units of perception, effects of syntactic roles, unit, organization and access to the mental lexicon, effects of context, as well as effects of repeated listening. A system for automatic recognition of continuous speech is then proposed on the basis of these results and considerations, utilizing both words and syllables as units of recognition, and using both local and global coherence in reducing word candidates as well as modifying or renewing word and syllable templates.

In a text-to-speech conversion system, the prosodic features of speech should be synthesized strictly by rules. In order to construct a set of rules to generate the symbols for the prosodic features, such as the pause duration, the phrase command magnitude, and the accent command amplitude, speech samples of Japanese texts were analyzed quantitatively using the model for fundamental frequency contour generation. The analysis made clear the relationships between the prosodic features and the linguistic information, such as the accent type of a word, the syntactic structure of a sentence, and the discourse structure of a text. Based on these relationships, rules were constructed which generate the prosodic symbols from the linguistic information. Text-to-speech conversion was conducted using these rules. The naturalness of the intonation of the synthesized speech indicated the validity of the rules.

A method is described for the accurate extraction of formant trajectories from utterances containing voiced stop consonants, and for the determination of the formant loci from the trajectories. Analysis of CV and VCV utterances revealed that the loci of the second, third and fourth formants are dependent on the vowel following the consonant, but can provide sufficient information for the separation of [b], [d], and [g] in most cases. The rise time constant of the short-time power is found to supplement formant loci information in cases where the latter is not sufficient. It is shown that correct recognition of the three voiced stop consonants from a single speaker is possible by first identifying the following vowel and by using a set of context-dependent linear discriminant functions in a 4-dimensional space of the three formant loci and the rise time constant.

A system is proposed for automatic speech recognition using syllable templates. In this system, input speech signal is analyzed and matched against syllable templates and converted into parameters characterizing each candidate syllable. Word recognition is based on an overall likelihood measure calculated for each item stored in the lexicon. A method is also developed for the optimization of syllable templates. The validity of the proposed method was tested in a preliminary recognition experiment in which a lexicon consisting of 1000 city names was used to recognize utterances of 100 city names by a female speaker. The rate of correct recognition was 96.5%.

Announcements and narrations broadcast over radio and television have been serving in recent years as the most effective means for spreading the spoken version of the common Japanese, especially in regards to intonation. No formal studies, however, seem to have been conducted on the objective analysis and formulation of the rules of intonation of the common Japanese. The present work investigates the acoustical and perceptual characteristics of the intonation of the common Japanese used by radio and television announcers, and attempts to derive rules of intonation for efficient transmission of spoken messages. The results are expected to be of value both for high-quality speech synthesis and for scientific and objective teaching of the spoken version of the common Japanese.

The causes of errors in conventional pitch extraction methods can be classified into two categories: 1) extrinsic factors that come from the analysis methods, such as imperfect signal representation due to inappropriate shape, width, and placement of the analysis window, etc., and 2) intrinsic factors that reside in the speech signal itself, such as the occurrence of a strong harmonic component or a sub-harmonic component, etc. In fact, fixed frame size and frame shift, adopted in most of the conventional systems, are responsible for a considerable part of their gross pitch errors. In this paper we combine the use of running waveform analysis, an exponential analysis window, and a variable frame shift to cope with errors of the first category. Various new methods are also introduced to cope with causes of errors of the second category. The validity of the proposed methods is confirmed by experiments using speech materials from both male and female speakers.

Speech analysis for high quality speech synthesis or high accuracy speech recognition requires realistic models not only for the vocal tract but also for the voice source. In the present paper, we investigate models for the glottal volume velocity waveform. Previously proposed models are reviewed and classified according to their level of elaboration in expressing the glottal characteristics. A new model is then proposed which possesses all the important features of previously proposed models. A method is also described for simultaneously estimating the glottal source and vocal: tract parameters. Using this method, evaluation of glottal model parameters is carried out on real speech by varying the number of parameters in the proposed model. The results indicate the importance of detailed modeling of the period of glottal closure for accurate analysis.

In spite of its extensive use, speech analysis based on linear prediction (LP) is liable to various causes of inaccuracy. This paper presents a novel approach to improve the accuracy in the estimation of the voiced speech production model based on the LP method. The presented method uses interpolation between spectral points which are least influenced by artifacts in the spectral analysis and by noise in the signal. We show, on analyses of both synthetic and natural speech, that the averaged parabolic approximation between harmonic peaks of voiced speech spectrum reduces the sensitivity of the LP analysis to changes in the fundamental frequency Fo and to noise. The method is well suited for combination with the Spectral Transform LP method, previously proposed by the authors [1].

Although it has been well known that prosody plays an important role both in the intelligibility and in the naturalness of speech, the process of generating natural prosody from linguistic information has not been fully understood. The authors first define units of prosody of spoken Japanese on the basis of analysis of fundamental frequency contours. Prosodic words are defined by the presence of an accent component, while prosodic phrases and clauses are defined by the presence/absence of a pause and resetting/addition of a phrase component. It is shown that the accent components, representing the information concerning lexical word accent, are modified systematically by the syntactic and the discourse information. Classifications of prosodic boundaries are also presented, and the relationship between these two kinds of boundary is described. >

On the basis of the short-time relative speech rate defined by the authors, this paper examines the optimum width of the smoothing window by perceptual experiments on the naturalness of re-synthesized speech. With the optimum window of 270 ms, relative speech rates are obtained both for 'fast' and 'slow' utterances of the same sentence, using an utterance produced at a 'normal' speech rate. The averaged results show that the speech rate control function for an utterance can be approximately decomposed into a global component for each sentence and local components for each bunsetsu and each major syntactic boundary. Based on these results, a scheme is presented for controlling the local speech rate of a reference utterance to obtain a synthetic utterance of an arbitrary global speech rate.

The process of generating an F/sub 0/ contour from a small number of linguistically meaningful parameters, has been modeled quite accurately, and the model has been used extensively in speech synthesis. The paper deals with the inverse problem, i.e., that of extracting the model parameters from a given contour, which can only be solved by successive approximation. It presents a method for deriving a first-order approximation to a given F/sub 0/ contour from the linguistic information of the utterance, and refining the approximation by analysis-by-synthesis. The validity of the method has been confirmed experimentally.

The process of generating an F/sub 0/ contour from a small number of linguistically meaningful parameters, has been modeled quite accurately, and the model has been used extensively in speech synthesis. The study deals with the inverse problem, i.e., that of extracting the model parameters from a given contour, which can only be solved by successive approximation. This paper presents a method for deriving a first-order approximation to a given F/sub 0/ contour from the linguistic information of the utterance, and refining the approximation by analysis-by-synthesis. The validity of the method has been confirmed experimentally.

A text-to-speech conversion system for Japanese, developed for the purpose of producing high-quality speech output, is presented. It consists of four processing stages: (1) linguistic processing, (2) phonological processing, (3) control parameter generation, and (4) speech waveform generation. An overview of the whole system is presented. The innovations introduced into the second and the fourth stages, i.e. rules for generating prosodic symbols from the linguistic information in the second stage and the configuration of a new type of terminal analog speech synthesizer designed as the fourth stage, are described. The validity of the approach is confirmed by the improvements both in the prosodic and in the segmental qualities of synthesized speech. >

A method of path-limitation is proposed for the efficient DP matching of spoken words. Results of a preliminary recognition experiment show that this method can reduce the recognition errors as well. A method for effective speaker normalization is proposed. This method is based on the transformation of the relative positions of the five vowels of a given speaker in the parameter space. As a result of transformation, the dimension of the parameter space is reduced by a factor of four, yielding a great advantage in the computation time for matching. A study on the realization of a spoken-word recognition device using systolic array processors is described. A cyclic configuration that permits real-time matching of 600 templates with a single LSI chip is proposed. >

Accentuation serves to express both the discrete information concerning the accent type of a prosodic word and the continuous information concerning its prominence. The paper examines the latter aspect of accentuation using recorded radio news read by announcers. The amplitude of the accent command was extracted from an F/sub 0/ contour and used as an index for the level of accentuation. Statistical analysis of the accent command amplitude confirmed the difference between accented and unaccented types. Further analysis of the relationship between amplitudes of two adjoining accent commands also revealed a marked difference in characteristics of these two types.

Conventional speech analysis methods based on linear prediction often fail to separate and estimate the source and vocal-tract characteristics, especially in the case of voiced sounds, because of oversimplified assumptions regarding the voice source. We have already proposed a model that is capable of expressing a wide range of voice source characteristics, and demonstrated that source and vocal-tract parameters can be well separated and correctly estimated, for vowel and vowel-like sounds, by combining the proposed source model with the linear predictive analysis. The present paper extends our approach to apply to a wider variety of speech sounds including nasal vowels and nasal consonants, by combining the proposed source model with the ARMA analysis. The validity of the system was demonstrated by analysis of synthetic and natural speech.

In Swedish, the fundamental frequency contour (F/sub 0/ contour) is known to be the main acoustic feature for word accent and sentence intonation. The authors propose a model for the generation process of F/sub 0/ contours of Swedish, based on an extension of the model for Japanese. As the input to this model, two kinds of command are assumed: phrase commands, which are always positive impulses, and accent commands, which are stepwise functions of both polarities. Analysis-by-synthesis of F/sub 0/ contours of both isolated words and sentences, uttered by two native speakers from the Stockholm region, indicated that the model can always generate very close approximations to observed F/sub 0/ contours, and that the extracted parameters are systematically related to the underlying lexical word accent, syntactic structure, and focus. This demonstrates the basic validity of the model and its utility in speech synthesis. >

Although pitch extraction schemes based on the short-time autocorrelation function offer reliable results for most of the speech signal, the autocorrelation peaks indicating fundamental period fluctuate with frame position, causing occasional pitch extraction errors. In order to reduce these errors, a scheme using a new definition of normalized short-time autocorrelation function is proposed. One of the major advantages of the definition over conventional ones is that the frame length changes in proportion to the time lag, and, therefore, the input speech can be analyzed without any knowledge of the fundamental frequency range of the speaker. Two methods are proposed for the normalization, and the one compensating for variations of the short-time power of the waveform is shown to offer better results. A system for pitch extraction is constructed on a workstation, and the validity of the proposed scheme is demonstrated by experiments using the connected speech of male and female announcers. >

Two analysis methods are proposed to estimate the orders as well as the parameter values of an autoregressive moving-average (ARMA) process with a white Gaussian noise or a train of impulses as input. This process is adopted as a model for the process of speech production. The estimation is accomplished directly from the signal waveform using recursive formulas and is optimum in the sense that the variance of errors in the estimated waveform is minimized. This method is less vulnerable to fluctuations in the fundamental period of the source than the previously proposed methods based on the auto-correlation function. In order to trace the trajectories of the poles and zeros in the vocal tract transfer function, it is required to develop a method for adapting to the temporal variations of the acoustic characteristics of speech. A method is also proposed for adaptive control of the analysis interval based on the detection of convergence of the estimated values of the parameters. The validity of the proposed methods is tested by analysis of both synthetic and natural speech sounds.

SpectraI analysis of the Japanese vowels shows that the five vowels /a/, /e/, /i/, /o/, and /u/ of a single speaker can well be separated by their first and second formant frequencies (F 1 and F 2 ). Considerable amount of overlap is observed, however, when vowels of many speakers are plotted in the F 1 -F 2 plane, which can be ascribed mainly to differences in the size and shape of the vocal tract. A normalizing process, based presumably on higher formant frequencies, is expected in the identification of these vowels. It is not dear, however, whether concurrent changes of pitch and higher formants are necessary in the normalization process. This paper presents a method for evaluating the roles of these parameters and describes the results obtained. Perceptual boundaries between a pair of vowels, which share approximately the same ratio of F 2 to F 1 , are defined in the F 1 F 2 plane, using synthetic vowels generated by a terminal analog synthesizer. The importance of pitch and higher formants, is then evaluated by the extent to which their changes affect these boundaries. The results of listening tests show that, for ordinary buzz-excited vowels, neither pitch nor higher formants alone are sufficient for perceptual normalization, and the combined changes in pitch and higher formants are necessary to counteract the changes in F 1 and F 2 . For noise-excited vowels, on the other hand, the roles of higher formants are as important as the combined roles of pitch and higher formants in buzz-excited vowels.

Procedures are described for reducing the speech wave to a specification in terms of the time‐varying vocal‐tract resonances and source characteristics. The basic method, which has been called analysis by synthesis, involves the comparison of speech spectra with a series of spectra that are synthesized within the analyzer. Each comparison spectrum is generated according to a set of rules based on an acoustical theory of speech production. The result of the analysis of each input spectrum is a set of parameters that describes the synthesized spectrum providing the best match. In one version of the method convergence, towards the best match is controlled by the experimenter; in another version convergence to a match is accomplished automatically without the intervention of the experimenter. All the operations have been programmed on a general‐purpose digital computer and have been applied to the analysis of vowels and some consonants. The advantages of the analysis techniques are discussed.

The principle of analysis‐by‐synthesis, described in earlier papers, is applied to the measurement of the formant frequencies of the common vowels of American English. The vowels measured constitute segments of bisyllabic nonsense utterances recorded by three adult male talkers. The set of materials is designed to provide samples of stressed vowels preceded and followed by a variety of consonants. The vowels in question occur in consonant‐vowel‐consonant syllables in which both consonants represent the same phoneme of American English. Frequencies of the first three formants are measured throughout the vowel for spectral samples separated by approximately 8 msec. Effects on formant frequencies attributable to temporal location of sample, consonant environment, and individual talkers are discussed. For central locations of the samples, the formant frequencies differ systematically from published data derived from more restricted consonantal environments. Phenomena relating to formant transitions and individual talker characteristics are discussed. [This work was supported in part by the U. S. Army (Signal Corps), the U. S. Air Force (Office of Scientific Research, Air Research and Development Command), and the U. S. Navy (Office of Naval Research); and in part by the Air Force Cambridge Research Center.]