Paintable Computing Project Fall 2003 Virtual Self Assembly Turing substrate By way of introduction, I am the Capitan, Navigator, Oarsman, and Deck-swabbie on the paintable computing project here at MIT. And in my time with you, I would like to first put forth the notion of “Virtual Self-assembly”: - what it is - how far we’ve taken it - and why anyone should care. Summarizing it’s relevance, self-assembly in the virtual domain offers a promising avenue for overcoming the really gnawing complexity limitations of self-assembly in the physical domain. A dense mesh of spatially distributed locally interconnected Turing machines Virtual Self-assembly Spontaneous aggregation of spatially mobile fragments of code with state.

Paintable Computing Project Fall 2003 Turing Substrate - What Turing substrate 1) A dense irregular mesh of individually programmable Turing machines. 2) a 3D volume populated by logic-enabled nodes which are computationally universal. See also: computational fabric. Virtual self-assembly is predicated on -- and motivated by -- the notion of a “Turing substrate” . We use this name to describe a 3D volume populated my a dense irregular rmesh of logic-enabled nodes that are computationally universal This abstraction is a straightforward extension of work that models nature as an irregular mash of fine grain computing elements, each with task-specific physical hardware running a limited internalized procedure. For errant chip designers like myself, the notion of a Turing Substrate is simply an admission that today, you can fit a 32-bit Palm-class computer and a quarter meg of SRAM into 2 square mm of die.

Paintable Computing Project Fall 2003 Turing Substrate - Who Turing substrate 1) A dense irregular mesh of individually programmable Turing machines. 2) a 3D volume populated by logic-enabled nodes which are computationslly universal. See also: computational fabric. For a number of groups throughout the world, this is a core concept if not a term of art. Our work on paint specifically targets traditional silicon processing elements in sand-sized packages separated by millimeter length scales. For us, the vision is scoop out the dust, pour it into a paint-like goo, plaster it down, expose it to power, and it does the reset. A belief shared throughout this community has been that the hardware development was mostly evolutionary. But that the software was going to send us back to the drawing board. Smart Dust (U. C. Berkeley) Amorphous Computing (MIT LCS/AI-lab) Paintable Computing Project (CBA)

Paintable Computing Project Fall 2003 Virtual Self-assembly Virtual Self-assembly 1) The undirected reassembly of a process from randomly distributed fragments of code with state. 2) The spontaneous aggregation of a distributed process from mobile process fragments migrating within a computational medium. See also: Wack-o F1() Xn F1() Xn F1() Xn Xn Xn F1() F1()

Paintable Computing Project Fall 2003 Virtual Self-assembly Virtual Self-assembly 1) The undirected reassembly of a process from randomly distributed fragments of code with state. 2) The spontaneous aggregation of a distributed process from mobile process fragments migrating within a computational medium. See also: Wack-o F1() Xn F1() Xn F1() Xn Xn 1110010101 0110101010 1110010101 Xn F1() 1010010101 1110010101 1110010101 1110010101 0110101010 1001010100 F1() 1010100010 1110010101 1110010101

Paintable Computing Project Fall 2003 Virtual Self-assembly Virtual Self-assembly 1) The undirected reassembly of a process from randomly distributed fragments of code with state. 2) The spontaneous aggregation of a distributed process from mobile process fragments migrating within a computational medium. See also: Wack-o F1() Xn F1() Xn F1() 1110010101 Xn Xn 0110101010 1110010101 1010010101 Xn F1() 1110010101 1110010101 1110010101 0110101010 1001010100 F1() 1010100010 1110010101 1110010101

Paintable Computing Project Fall 2003 Virtual Self-assembly Virtual Self-assembly 1) The undirected reassembly of a process from randomly distributed fragments of code with state. 2) The spontaneous aggregation of a distributed process from mobile process fragments migrating within a computational medium. See also: Wack-o F1() Xn F1() F1() Xn F1() Xn Xn Xn F1() F1()

Paintable Computing Project Fall 2003 Tessellation via Thermodynamic (-like) relaxation

Paintable Computing Project Fall 2003 Distributed Graphics Controller

Early Applications Media Streaming Holistic Data Storage Surface Bus Some other applications we have demonstrated in simulation: Streaming Media -- where packetized media is modeled as a diffusive “data gas” that is streamed through a single port, randomizes its position for storage, and re-assembles itself into a serial order for output Holistic Data Storage builds on this idea by adding signal processing to construct holistic representations for image storage – here, the data gas reacts enzymatically with pfrags that perform space-frequency transformation on the packets. The result that undersampling of the packets results in reconstructed images that are undersampled in spectrum -- much like a hologram. In Surface Bus, objects placed on the periphery of a 2D surface diffuse into the 2D ensemble small process fragments that replicate and self organize into a dynamic token ring. And finally, Image Segmentation is realized as the competition among process fragments, each selective for image characteristics such as sky, water, trees, and sand Holistic Data Storage Surface Bus Image Segmentation

Paintable Computing Project Fall 2003 Why Virtual Self-assembly? = ? Turing Substrate + Sensing/Actuation Universal Self-assembler Beyond individual applications lurks the real questions of how this technique relates to the fundamentals of a budding field. I see three points worth raising: 1) The first, a question. Does a threshold number of universal computers, arranged in a dense irregular mesh, and each with some capacity to sense and actuate the physical world, beget an universal assembler. I am totally out of my water make or refute the claim. But the question begs an answer. And I am grateful for anyone’s input. 2) Back on the shop floor, The combination of a Turing substrate whose nodes are stamped out by millions in mature wafer foundries, together with mobile code whose replication machinery is a simple digital copy, seems an interesting alternative to von-Neuman’s original abstraction of machine of potentially unbounded complexity replicating itself. Copy

Paintable Computing Project Fall 2003 Why Virtual Self-assembly? Defect likelihood as limitation to growth Stoichiometric limits on component placement 3) Finally, virtually self-assembly holds out the promise of overcoming the gnawing problems of contemporary techniques for self-assembly. will likely be most profitably used as an adjunct to material self-assembly. On deck for my work is a iterative sequencing of material and virtual self-assembly. With material resources assembling into coarse aggregates followed by detailed functional assignment in the virtual domain. At which point the sequence repeats using the previous result as a dynamic scaffold. Virtual Self-assembly Stage 2 Stage 4 Material Self-assembly Stage 1 Stage 3

Paintable Computing Project Fall 2003 Title F1()