amorphous computing
Download
Skip this Video
Download Presentation
AMORPHOUS COMPUTING

Loading in 2 Seconds...

play fullscreen
1 / 37

AMORPHOUS COMPUTING - PowerPoint PPT Presentation


  • 134 Views
  • Uploaded on

AMORPHOUS COMPUTING. Harold Abelson, Thomas F. Knight, and Gerald Jay Sussman Massachusetts Institute of Technology. A scientific and technological effort to identify.

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'AMORPHOUS COMPUTING' - amos


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
amorphous computing

AMORPHOUS COMPUTING

Harold Abelson, Thomas F. Knight,

and Gerald Jay Sussman

Massachusetts Institute of Technology

DARPA Ultrascale Computing Program

a scientific and technological effort to identify
A scientific and technological effort to identify
  • methods for obtaining coherent behavior from the cooperation of large numbers of unreliable parts that are interconnected in unknown, irregular, and time-varying ways
  • techniques for instructing myriads of programmable entities to cooperate to achieve particular goals
  • engineering principles and languages that can be used to observe, control, organize, and exploit the behavior of programmable multitudes.

DARPA Ultrascale Computing Program

amorphous computing is
Amorphous computing is
  • Not emergent behavior
  • Not “pretty pictures”
  • It is engineering of mechanisms with prespecified, well-defined behavior

DARPA Ultrascale Computing Program

our model
Our model
  • Computing elements sprinkled on a surface or in a volume
  • Each can talk to a few nearby neighbors, but not reliably
  • Each has modest computing power and a modest amount of memory
  • The particles are not synchonized, nor are they regularly arranged.

DARPA Ultrascale Computing Program

properties of our model
Properties of our model
  • Local communication
    • locally disordered, but constrained by geometry
  • Too many components to individually program or even to name
  • Components are possibly faulty, sensitive to the environment, and may effect actions

DARPA Ultrascale Computing Program

example of an amorphous computing medium
Example of an Amorphous Computing Medium

An amorphous medium has independent computational particles, all identically programmed. Each particle is represented by a spot in the picture, and different states are shown by different colors.

DARPA Ultrascale Computing Program

why is this interesting
Why is this interesting?
  • Physically feasible at any scale
  • Forces robustness of design
  • Potentially extremely inexpensive
  • Provides the possibility of bulk computation
    • smart paints
    • smart gels
    • concrete by the Megaflops

DARPA Ultrascale Computing Program

how can we program amorphous stuff
How can we program amorphous stuff?
  • We can look to biology for organizational metaphors (although we do not try to duplicate actual biological mechanisms)
  • We’ll take examples from pattern formation to illustrate the point, and produce cartoon caricatures of biological morphogenesis

DARPA Ultrascale Computing Program

bifurcating tubes an example of emergent behavior
Bifurcating Tubes: an example of emergent behavior

This is an amorphous physical simulation of a weak membrane

bounding a pressure vessel. When a bulge appears, the membrane

thins and the bulge expands. (by Radhika Nagpal)

DARPA Ultrascale Computing Program

but suppose we wanted to make something with a precisely specified geometry
… But suppose we wanted to make something with a precisely specified geometry?

from Frank Netter

Atlas of Human Anatomy

DARPA Ultrascale Computing Program

local simd paradigm for programming differentiation and growth ron weiss
Local SIMD paradigm for programming differentiation and growth (Ron Weiss)
  • Each computing element’s state includes some binary markers. Each computing element’s program has many independent rules.
  • Rules are triggered when messages are received. A rule is applicable if a certain boolean combination of markers is satisfied.
  • When a rule is applied it may set markers and send further messages.
  • Messages have hop counts that determine how far they will diffuse.
  • Markers may have lifetimes after which they expire.

DARPA Ultrascale Computing Program

slide12
Differentiation

To make a “spine” the elements in an initial polarized tube must differentiate into bands of alternating C and D type segments.

DARPA Ultrascale Computing Program

a program for creating segments
A program for creating segments

(start

Crest

((send (make-seg C 1) 3)))

((make-seg seg-type seg-index)

(and Tube (not C) (not D))

((set seg-type)

(set seg-index)

(send created 3)))

(((make-seg) (= 0))

Tube

((set Bottom)))

(((make-seg) (> 0))

Tube

((unset Bottom)))

(created

(or C D)

((set Waiting 10)))

(* (and Bottom C 1 (Waiting (= 0)))

((send (make-seg D 1) 3)))

(* (and Bottom D 1 (Waiting (= 0)))

((send (make-seg C 2) 3)))

(* (and Bottom C 2 (Waiting (= 0)))

((send (make-seg D 2) 3)))

(* (and Bottom D 2 (Waiting (= 0)))

((send (make-seg C 3) 3)))

DARPA Ultrascale Computing Program

slide14
Segments can grow by invading neighboring

segments, thereby stimulating them to grow

DARPA Ultrascale Computing Program

an experimental prototype for amorphous computing
An experimental prototype for amorphous computing

Chris Hanson

Don Allen

Darren Schmidt

Bei Wang

Chris Terman

Tom Knight

DARPA Ultrascale Computing Program

computational particles communicate using spread spectrum transceivers
Computational particlescommunicate using spread-spectrum transceivers.

DARPA Ultrascale Computing Program

test setup for transceiver
Test setup for transceiver

DARPA Ultrascale Computing Program

test signals from transceiver
Test signals from transceiver

DARPA Ultrascale Computing Program

some general tools for building structure in amorphous systems
Some general tools for building structure in amorphous systems
  • Wave propagation
  • Coordinate refinement
  • Activation/inhibition

DARPA Ultrascale Computing Program

slide21
Crude local coordinate systems can be constructed by intersecting countdown waves radiating from several loci.

DARPA Ultrascale Computing Program

cleverness makes better coordinates
Cleverness makes better Coordinates

Laplace’s equation can be used to interpolate from a boundary condition. One can approximately solve Laplace’s equation on an amorphous computer by successive averaging.

DARPA Ultrascale Computing Program

local coordinate systems can be combined
Local coordinate systems can be combined.

A global picture can be constructed by working out the mappings at the overlaps among the local coordinate systems.

DARPA Ultrascale Computing Program

slide25
By combining these ideas, we can obtain detailed topological control, and embody the control mechanisms in a language…

Cartoon caricature of “differentiation” to build a chain of “CMOS inverters” (by Daniel Coore)

DARPA Ultrascale Computing Program

a botanical metaphor
A botanical metaphor

We organize the process in terms of “growing points.”

They make structures that exhibit “tropisms” toward particular “chemical gradients.”

The growing points may lay down materials.

Materials may secrete pheromones that attract or repel other growing points.

Growing points may split, die off, or join.

Support for this abstraction may be programmed as a uniform state machine in each computational particle.

DARPA Ultrascale Computing Program

start with vdd vss and a poly contact
Start with Vdd, Vss, and a Poly Contact

DARPA Ultrascale Computing Program

slide28
The poly contact sprouts a growing point that bifurcates and then grows toward the pheromones secreted by Vdd and Vss.

DARPA Ultrascale Computing Program

when the growing poly gets close to vdd and vss it is stopped by a short range inhibition
When the growing poly gets close to Vdd and Vss it is stopped by a short-range inhibition

DARPA Ultrascale Computing Program

slide30
The poly growing points die off, but first they sprout P and N transistor diffusion growing points, which grow toward Vdd and Vss, where they drop contacts.

DARPA Ultrascale Computing Program

slide31
The diffusions also grow toward each other. When they hit they form a new poly contact and poly growing point.

DARPA Ultrascale Computing Program

this process repeats to make an arbitrarily long chain of ugly but topologically correct inverters
This process repeats to make an arbitrarily long chain of ugly, but topologically correct inverters.

This demonstrates totally local control of precision topology.

DARPA Ultrascale Computing Program

slide34
This very parallel process can be described using a serial process metaphor. The growing points provide a serial locus of control, even though the implementation is in terms of a uniform state machine in each computational particle

(define-growing-point ((poly from-input-contact) Q-id)

(material poly)

(tropism constant Vdd-long Vss-long)

(initialize lifetime 5)

(secrete (poly-short inhibitor) Q-id)

(when (= lifetime 0)

(start-growing-point (poly up) Q-id)

(start-growing-point (poly down) Q-id)

(terminate)))

DARPA Ultrascale Computing Program

the growing points created by the poly contact growing point have independent existence
The growing points created by the poly contact growing point have independent existence.

(define-growing-point ((poly up) Q-id)

(material poly)

(secrete (poly-short inhibitor) Q-id)

(tropism increasing Vdd-long)

(when (sensing? Vdd-short)

(start-growing-point (poly p-fet) Q-id)

(terminate)))

(define-growing-point ((poly p-fet) Q-id)

(material poly)

(initialize lifetime 10)

(secrete (poly-short inhibitor) Q-id)

(tropism constant Vdd-short)

(when (= lifetime 1)

(start-growing-point (diffusion p-fet Vdd))

(start-growing-point (diffusion p-fet contact)

Q-id)))

DARPA Ultrascale Computing Program

growing points can join as well as split
Growing points can join as well as split.

(define-growing-point ((diffusion p-fet Vdd) Q-id)

(material diffusion p-type)

(secrete (diffusion-short inhibitor) Q-id)

(tropism increasing Vdd-short)

(when (is-type? Vdd)

(drop-metal-diffusion-contact)

(terminate)))

(define-growing-point ((diffusion p-fet contact) Q-id)

(material diffusion p-type)

(secrete (diffusion-short inhibitor) Q-id)

(secrete (n-diffusion-attract attractor) Q-id)

(tropism increasing p-diffusion-attract Q-id)

(when (is-type? n-diffusion)

(drop-poly-diffusion-contact)

(start-growing-point (poly from-input-contact)

(new-id))

(terminate)))

DARPA Ultrascale Computing Program

the challenge of amorphous computing
The Challenge of Amorphous Computing
  • To reliably obtain a desired behavior by engineering the cooperation of many parts, without assuming any precision interconnect or precision geometrical arrangement of the parts.
  • To invent the computational substrate that can support this kind of engineering.

DARPA Ultrascale Computing Program

ad