home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!haven.umd.edu!darwin.sura.net!zaphod.mps.ohio-state.edu!uwm.edu!lll-winken!tazdevil!henrik
- From: henrik@mpci.llnl.gov (Henrik Klagges)
- Newsgroups: comp.ai.neural-nets
- Subject: Re: Intel 8017ONX analogue neural network chip. HELP!!!
- Keywords: hardware 8017ONX analogue neural network trainable
- Message-ID: <?.714767138@tazdevil>
- Date: 25 Aug 92 18:25:38 GMT
- References: <1992Aug24.011446.12790@usage.csd.unsw.OZ.AU>
- Sender: usenet@lll-winken.LLNL.GOV
- Lines: 21
- Nntp-Posting-Host: tazdevil.llnl.gov
-
- srms@newt.phys.unsw.edu.au (Sergio Sandler) writes:
-
- >Somebody told me that INTEL is selling an electronically trainable
- >analogue neural network chip no. 8017ONX, having 64 neurons. If this
- >information is correct, could somebody tell how could I get that chip,
-
- I looked at the specs of the 80170 about a year ago. As far as I remem-
- ber, it is NOT trainable in the classical fashion. If I remember correctly,
- it is weight-burn in and forget - so it essentially always runs in production
- mode. (they may have an eeprom-like reprogramming feature, but if, it is not
- fast enough for online learning).
- Apart of that, the numerical precision obtainable with the resistor array is
- not too exciting. Last I heard was that manufacturing tolerances allowed for
- a resolution of 256 values at best for stored parameters (e.g., weights), and
- monotonicity couldn't be guaranteed.
-
- Cheers, Henrik
- Massively parallel group at Lawrence Livermore
- IBM Research division physics group Munich
-
- PS: May very well be possible that they upgraded that chip by now.
-