~e; tuning analog circuits

From bc <human@electronetwork.org>
Date Mon, 25 Feb 2002 16:20:11 -0600

  [i would not have guessed that analog circuits could be better, given
  the doctrine (and resulting dogma) in the digitization of the world. it
  always was odd, that audiophiles kept their vacuum tube equipment
  running "because it sounds richer" or something such. maybe music-
  ians have always already known this, given soundwaves and samples
  brought about by the digital segmentation of a potential smooth wave
  of aural or other events. audiovisual being analog... to what. wonder.]

Upside of Downsizing Analog Chips


By Manny Frishberg

2:00 a.m. Feb. 20, 2002 PST

SEATTLE -- Portable phones that fit in your ear or Dick Tracy's
two-way wrist TV are still science fiction for a simple reason: Until 
now, Moore's Law has only applied to digital circuits.

Intel co-founder Gordon Moore's prediction that the number of
transistors on a chip would double every 18 months has pretty much 
borne out, and the digital revolution has spawned with it. Shrinking 
the size of the circuits has led us from chips with a few thousand 
transistors each, in the 1970s, to today's super-speedy processors 
and billion-byte DRAMs.

But while smaller equals faster in digital technology, the same has
not been the case in analog.

"With analog circuits that process real-world signals, when you scale
the transistors smaller and smaller, the performance gets worse and 
worse," says Bill Colleran, CEO of a Seattle company named Impinj, 
which recently received the 2002 Best New Technology Award from WSA 
(formerly the Washington Software Alliance) for its self-adaptive 
silicon technology.

"So you have this dichotomy where digital wants things to get smaller
and smaller and everything gets better, and analog gets worse and worse."

Analog devices are found in cell phones, MP3 players, cable modems and
most communication devices. While digital chips are nearly ideal for 
storing and manipulating data -- as ones and zeros -- to do 
everything from word processing and spreadsheet calculations to 
playing games, they are ill-suited for communicating. Dealing with 
sound or radio and TV signals requires analog technology, which works 
with information in the form of waves of various sizes and shapes.

Everything with a microphone, a speaker or a camera uses analog
signals that need to be translated into digital form and back again.

Intel and other chip manufacturers are attempting to address the
problem of building mixed analog and digital "systems-on-chip" by 
migrating to new production processes. The use of exotic materials 
such as gallium arsenide is very expensive and requires massive 
retooling of their labs.

But Impinj has found a way to make the analog devices with silicon,
employing the same CMOS technology currently used for making digital 
chips and fine-tuning them after they are produced. The result is 
analog devices such as digital-to-analog processors that can be 
scaled down to tiny sizes and work better than the current generation 
of analog chips.

Producing millions of transistors on dozens of chips from a single
silicon wafer is fine for digital circuits, which only need to be 
accurate enough to discern between a one and a zero, Colleran said. 
But with analog, you have to process signals over a full range of 
values. And if the transistors don't behave exactly the way they're 
supposed to, the ability to process those signals correctly is 

Impinj likens its analog circuits to a dimmer on a light switch, which
can be set to different levels, as opposed to a digital circuit that 
can only be turned on or off. Passing different electric currents 
through the circuit fine tunes the CMOS transistors after they have 
been manufactured.

The process is modeled on how the human brain adjusts the nerve cells.
Called "self-adaptive silicon" technology, it can monitor the chip's 
functioning and reset it to adapt to changes in temperature or 
battery power.

Co-founders Carver Mead, who pioneered the technology behind today's
super powerful chips, and his former student, Christopher Diorio, 
developed the process while at Cal Tech in the 1990s. Impinj has 
bartered royalty-free licenses for more than a dozen basic patents 
from Cal Tech in return for an equity stake in the company.

"Our strength is doing analog in an integrated fashion," Mead says, by
getting "analog to coexist with digital in a straight-forward, plain 
vanilla CMOS process, which is what all the chips in the world are 
made out of. We can go in post production and get them to match. That 
enables us to modify the characteristics of individual transistors 
after they've been manufactured."

This ability grew out of research that Mead and Diorio conducted at
Cal Tech in the 1990s.

"As digital circuits shrink to ever smaller feature sizes, the value
in the analog portion, that doesn't scale well, continues to rise," 
said Ed Lazowska, computer science professor at the University of 
Washington where Diorio is now an associate professor. "Impinj's 
analog circuits are simple to design because they self-tune, are 
small because the transistors themselves compensate for mismatch and 
degradations, and (because they) learn from their inputs."

"The future of self-adaptive silicon technology is broad," Lazowska
said. "It can improve the battery life in wireless systems, enable 
low-power adaptive sensors, and allow silicon chips that learn from 

Copyright (C) 1994-2002 Wired Digital Inc. All rights reserved.

[fair-enough-use, .edu 2oo2 analog|digital epiphany]

  the electronetwork-list
  electromagnetism / infrastructure / civilization