T. Lehmann, Hardware Learning in Analogue VLSI Neural Networks. PhD thesis, Electronics Institute, Technical University of Denmark, Lyngby, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....accuracy of digital hardware is perhaps only limited by bit resolution, integral to analogue circuitry are hardware errors. This section describes a number of the common sources of this inaccuracy at an integration level and also at a circuit and system level. For a more detailed discussion see [20] and the included references. At an integration level process variation is the cause of what can be a significant amount of hardware error. One component of this is local process variation which leads to inevitable random fluctuation in all parameters. Local hardware errors can be due to, for ....

....In analogue VLSI this noise can add to the other sources discussed here as hardware errors. Therefore at an integration level errors occur that are fundamental to analogue circuits and although some steps can be taken to minimise the effects there will always be some level of inaccuracy. Lehmann [20] calculates that analogue computational hardware is typically limited to a relative precision of about 1 . In the next sections errors due to decisions made at a higher level are considered. When designing a circuit in analogue VLSI there will inevitably be a certain level of hardware error. Some ....

T. Lehmann, Hardware learning in analogue VLSI neural networks, PhD thesis, Technical University of Denmark, 1994.

.... In a paper discussing a particular analogue VLSI implementation of a neural network architecture we present results that show that hardware faults, in the specific implementation and we suggest more generally, can be modelled as weight perturbations [4] Such errors have many sources [5] and the repetitive structure of neural network architectures and the requirement of weight storage will exacerbate the problem. This paper therefore concentrates on this form of error small perturbations rather than stuck at faults. In this paper we propose the use of augmented networks where ....

LEHMANN, T.: `Hardware learning in analogue VLSI neural networks'. PhD Dissertation, Technical University of Denmark, 1994.

....model is unrealistic, at least for most current implementations of neural networks. Here we consider some of the common areas of hardware error and substantiate the above statement proposing our own small perturbation fault model. For a more detailed discussion on analog VLSI implementations see [8] and [7] or for the digital case [5] Errors occur at many levels of hardware implementation. For the analog VLSI case Lehmann [8] suggests that computational hardware is inherently limited to a relative precision of about 1 . This will be the result of the combination of many factors, such as ....

....hardware error and substantiate the above statement proposing our own small perturbation fault model. For a more detailed discussion on analog VLSI implementations see [8] and [7] or for the digital case [5] Errors occur at many levels of hardware implementation. For the analog VLSI case Lehmann [8] suggests that computational hardware is inherently limited to a relative precision of about 1 . This will be the result of the combination of many factors, such as process variation leading to component mismatch and changes in threshold voltages between chips or even of components on the same ....

T. Lehmann. Hardware learning in analogue VLSI neural networks. PhD thesis, Technical University of Denmark, 1994.

....and thus form a poor criterion against which to assess fault tolerance. Here we consider some of the more common areas of hardware error and substantiate the above statements proposing our own small perturbation fault model. For a more detailed discussion on analog VLSI implementations see Lehmann (1994) and Edwards and Murray (1997) or for the digital case, Edwards and Murray (1996a) Errors occur at many levels of hardware implementation. In analog VLSI it has been suggested (Lehmann 1994) that computational hardware is inherently limited to a relative precision of about 1 . This will result ....

....our own small perturbation fault model. For a more detailed discussion on analog VLSI implementations see Lehmann (1994) and Edwards and Murray (1997) or for the digital case, Edwards and Murray (1996a) Errors occur at many levels of hardware implementation. In analog VLSI it has been suggested (Lehmann 1994) that computational hardware is inherently limited to a relative precision of about 1 . This will result from a combination of factors, such as process variation. This leads to component mismatch and changes in threshold voltages and other parameters between chips or on the same chip. At a higher ....

Lehmann, T. 1994. Hardware learning in analogue VLSI neural networks. PhD thesis. Technical University of Denmark.

....that programmable analogue circuits could have an impact on the semiconductor industry, similar to that of the digital systems during the eighties. 1.1. 1 Research in Edinburgh The author s initial research plans were to apply techniques, which were developed for artificial neural networks [5,6,7,8] (in particular stochastic neural networks [9, 10] to real world applications [11] The early literature survey outlined the problems of interfacing a neural network to the environment [12,13,14,15,16,17] The uneven distribution of signal energy [18,19] in the frequency domain, common in most ....

.... Comparators: 118] 119] 120] 121] Current Mode: 48] 49] 55] 56] 57] 59] 60] 61] 62] 63] 64] 65] 66] 67] 68] 69] Field Programmable Analogue Arrays: 51] 76] 77] 78] 79] 80] 82] 83] 84] 85] 86] 87] 88] 89] 90] 91] 95] 96] 97] 98] 99] 102] Literature: [8] [14] 19] 22] 35] 36] 37] 38] 39] 40] 41] 42] 43] 44] 46] 50] 58] Log Domain: 122] 123] 124] 125] 126] 127] 128] 129] 130] 131] 133] 134] 135] 140] 141] 142] Manuals: 73] 74] 75] 81] 92] 93] 94] 117] Matching: 69] 100] 105] 107] 108] 109] 110] 111] ....

T. Lehmann, Hardware Learning in Analogue VLSI Neural Networks. PhD thesis, Electronics Institute, Technical University of Denmark, Lyngby, 1994.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC