From pister Mon Jul 23 11:28:18 2001

Neal - 
 let me give you a status update on what we're doing on the Smart
Dust project.  That will give you some idea of how close we are to
what you're looking for.

 We are still hoping to have a semi-functional wireless sensor node
in a few cubic millimeters by August 29.  The current approach uses
two chips, one MEMS, one vanilla CMOS.
 The CMOS chip contains:
   * an 8 bit, 10k sample/sec ADC
   * a DC light level sensor
   * a 1 Mbps asynchronous optical receiver
   * a trimable 8 MHz oscillator
   * a digital controller (simple finite state machine)

 The MEMS chip contains: 
   * a multi-output power supply to run the CMOS chip (separate 1V supplies
	for the analog and digital portions, and a 2.5V supply for the 
	optical receiver).
   * a corner-cube retroreflector for transmitting low-bit-rate optical
	data
   * a single-axis accelerometer with its own power supply.

The MEMS chip is still in fab, but all of the components have been
built and tested separately, so other than the usual list of a few
hundred known and unknown problems that can kill the run, we should
have working parts in late august.

The CMOS chip has been fabbed (in a quarter micron CMOS process from
National Semiconductor).  The ADC works, burning about 2uW of power.
The light sensor works, zero power (it generates power actually).
The 8 MHz oscillator works, burning about 1uW of power.
The digital controller hasn't been tested yet.
The optical receiver, which worked on the last run, doesn't seem to be
working on this one.  It is detecting packets, but it's also looking 
at noise and thinking that it's packets, and if we use too much 
energy in the communication laser (more than about 100nW incident) it
shuts down entirely.  Since the controller mostly ignores the incoming'
packets anyway, we may still be able to show that we can sense and
transmit from the system.  The total volume will be around 4 cubic
millimeters (displaced volume, not convex hull), and the total mass 
will be around 10 milligrams.  Those numbers assume that we don't do
any thinning of the substrate.  All of the action is in the top 50 
microns or so of the two chips, and the substrates that they are on
are around 550 microns thick, so in principle we could reduce the volume
to well under 1mm^3 by just grinding the substrate down.  In practice
that may be difficult, I don't know yet.  In future runs the substrate
will be selectively removed as a part of the process.

The accelerometer output goes directly into the ADC in this design, 
without any voltage amplification.  As a result, the range is fixed by
the mechanical design.  We should be able to resolve something like
0.1 to 0.5 gravities on a full scale that is roughly 100 times that.

ksjp