1. Introduction

This emerging technology combines modular robotics, systems nanotechnology and

computer science to create the dynamic, 3-Dimensional display of electronic information known

as Claytronics.

1.1 Nanotechnology Nanotechnology is not a novel technology. It dates back to 1959, when it was first used.

Nanotechnology, shortened to "nanotech", is the study of the controlling of matter on an atomic

and molecular scale. Generally nanotechnology deals with structures sized between 1 to 100

nanometer in at least one dimension, and involve developing materials or devices within that

size.

Figure 1.1 Nano Tubes

Nanotechnology is very diverse, ranging from extensions of conventional device physics

to completely new approaches based upon molecular self-assembly, from developing new

materials with dimensions on the nanoscale to investigating whether we can directly control

matter on the atomic scale.

1.2 Computer Science:Computer science or computing science (sometimes abbreviated CS) is the study of the

theoretical foundations of information and computation, and of practical techniques for their

implementation and application in computer systems. It is frequently described as the systematic

study of algorithmic processes that create, describe, and transform information. Computer

science has many sub-fields; some, such as computer graphics, emphasize the computation of

specific results, while others, such as computational complexity theory, study the properties of

computational problems. Still others focus on the challenges in implementing computations. For

1

example, programming language theory studies approaches to describe computations, while

computer programming applies specific programming languages to solve specific computational

problems, and human-computer interaction focuses on the challenges in making computers and

computations useful, usable, and universally accessible to people.

The focus of computer science is more on understanding the properties of the programs

used to implement software such as games and web-browsers, and using that understanding to

create new programs or improve existing ones.

1.3 Modular Robots: Modular robots are autonomous kinematic machines with variable morphology. Beyond

conventional actuation, sensing and control typically found in fixed-morphology robots, self-

reconfiguring robots are also able to deliberately change their own shape by rearranging the

connectivity of their parts, in order to adapt to new circumstances, perform new tasks, or recover

from damage.

Figure 1.2 Modular robot

For example, a robot made of such components could assume a worm-like shape to move

through a narrow pipe, reassemble into something with spider-like legs to cross uneven terrain,

then form a third arbitrary object (like a ball or wheel that can spin itself) to move quickly over a

fairly flat terrain; it can also be used for making "fixed" objects, such as walls, shelters, or

buildings. In some cases this involves each module having 2 or more connectors for connecting

several together. They can contain electronics, sensors, computer processors, memory, and

power supplies; they can also contain actuators that are used for manipulating their location in

the environment and in relation with each other. A feature found in some cases is the ability of

the modules to automatically connect and disconnect themselves to and from each other, and to

form into many objects or perform many tasks moving or manipulating the environment.

2

2. Universal Intelligent Matter

Intelligent matter is any material in which individual molecules or supra-molecular

clusters function as agents to accomplish some purpose. Intelligent matter may be solid, liquid,

or gaseous, although liquids and membranes are perhaps most typical. Universally

programmable intelligent matter (UPIM) is made from a small set of molecular building blocks

that are universal in the sense that they can be rearranged to accomplish any purpose that can be

described by a computer program. In effect a computer program controls the behavior of the

material at the molecular level. In some applications the molecules self-assemble a desired

nanostructure by “computing” the structure and then becoming in-active. In other applications

the material remains active so that it can respond, at the molecular level, to its environment or to

other external conditions. An extreme case is when programmable supra-molecular clusters act

as autonomous agents to achieve some end.

Although materials may be engineered for specific purposes, we will get much greater

technological leverage by designing a “universal material” which, like a general- purpose

computer, can be “programmed for a wide range of applications. To accomplish this, we must

identify a set of molecular primitives that can he combined for widely varying purposes. The

existence of such molecular operations might seem highly unlikely, but there is suggestive

evidence that it may be possible to discover or synthesize them.

2.1 Applications

The possible applications of UPIM are, static or dynamic (or interactive). By a static

application we mean one in which the intelligent matter computes into an equilibrium state, and

is inactive thereafter. Therefore static applications are most often directed toward generating

some specialized material with a computationally defined nanostructure. On the other hand,

dynamic or interactive applications never terminate, but always remain ready to respond to their

environment in some specified way; they are the truly “smart” materials.

3

2.1.1 Static Applications

Programs are ideally suited to creating complex data structures which can be converted to

complex physical structures by means of UPIM. Networks, chains, tubes, spheres, fibers, and

quasi crystalline structures are all straightforward to compute. The network resulting from such a

computation will be composed of computational groups (e.g., S, K,A) as well as inert groups,

which are manipulated by the computation but do not affect it. Typically, in these applications

the computational phase will be followed by a chemical phase in which the computational groups

are replaced by substances appropriate to the application (a sort of “petrification”). In addition to

the examples already mentioned, such an approach could be used to synthesize membranes with

pores or channels of a specified size and arrangement (determined either deterministically by the

program or stochastically by molecular processes).

A number of applications are suggested by the requirements of implementing small,

autonomous robots. Some of these will be controlled by very dense analog neural net- works, but

to achieve densities comparable to mammalian cortex (15 million neurons per square cm., with

up to several hundreds of thousands of connections each), we will need to be able to grow

intricately branching dendritic trees at the nanoscale. Generation of such structures is

straightforward with UPIM (e.g., using L-systems). The sensor and effector organs of micro

robots will also require very fine structures, which UPIM can be programmed to generate.

Of course, we should not neglect the potential of UPIM to do conventional computation,

such as solving NP-complete problems by massively parallel computation. For example, we

might replicate many copies of a program to test a potential solution, then mix them in a reaction

vessel with structures representing possible solutions, and wait for equilibrium to determine

actual solutions. The advantage of our approach to this kind of search problem over others, such

as DNA computation, is that our nanoscale test molecules are programmable.

2.1.2 Dynamic Applications

Dynamic intelligent matter is interactive in the sense that it is continually monitoring its

environment and capable of responding according to its program. That is, it is in a state of

temporary equilibrium, which can be disrupted by changes in the environment, resulting in

further computation and behavior as the material seeks a new equilibrium.

4

For example, a membrane with channels, such as mentioned above, could be made active

by having the channels open or close in response to environmental conditions, including control

commands transmitted optically or chemically. The program located in each channel is simple: in

response to its sensor state it executes one or the other of two effectors, blocking the channel or

not. The sensor and the medium would determine whether the channel is sensitive to global

conditions (e.g.. overall chemical environment or ambient illumination) or to its local

environment (e.g., molecules or light in its immediate vicinity).

Similarly, unanchored or free-floating molecular clusters (e.g., in colloidal suspension)

may react to their environment and change their configuration, thus affecting physical properties

of the substance, such as viscosity or transparency. Or they might polymerize or depolymerize on

command.

Unanchored supramolecular networks might also operate as semiautonomous agents to

recognize molecules or molecular configurations, and act upon them in some intended way (e.g.

binding toxins or pollutants). However, such applications will require the agents to operate in a

medium that can supply the reactants needed for computation.

These sorts of active intelligent matter will find many applications in autonomous micro

robots. For example, active membranes can serve as sensory transducers, responding to

conditions in the environment and generating electrical, chemical, or other signals. They can also

be programmed to self-organize into structures capable of preprocessing the input (e.g., artificial

retinas or cochleas). Further, it is a simple modification of a membrane with channels to make a

membrane with cilia that flex on command. By means of local communication, the cilia may be

made to flex in coordinated patterns. Similarly we may fabricate artificial muscles, which

contract or relax by the coordinated action of microscopic fibers. UPIM may also provide a

systematic approach to self-repair of autonomous robots and other systems, since if a robot’s

“tissues” were created by computational processes, then they can remain potentially active, ready

to restore an equilibrium disrupted by damage. Less ambitiously, materials can be programmed

to signal damage or other abnormal conditions.

5

3. Programmable Matter

Programmable matter is the dynamic application of Universal Programmable Matter. In

the past 50 years, computers have shrunk from room-size mainframes to lightweight handhelds.

This fantastic miniaturization is primarily the result of high-volume nanoscale manufacturing.

While this technology has predominantly been applied to logic and memory, it’s now being used

to create advanced micro electromechanical systems using both top-down and bottom-up

processes.

One possible outcome of continued progress in high-volume nanoscale assembly is the

ability to inexpensively produce millimeter-scale units that integrate computing, sensing,

actuation, and locomotion mechanisms. A collection of such units can be viewed as a form of

programmable matter.

Programmable matter refers to a technology that will allow one to control and

manipulate three-dimensional physical artifacts (similar to how we already control and

manipulate two-dimensional images with computer graphics). In other words, programmable

matter will allow us to take a (big) step beyond virtual reality, to synthetic reality, an

environment in which all the objects in a user’s environment (including the ones inserted by the

computer) are physically realized. Note that the idea is not to transport objects nor is it to recreate

an objects chemical composition, but rather to create a physical artifact that will mimic the

shape, movement, visual appearance.

3.1 Synthetic reality

One application of an ensemble, comprised of millions of cooperating robot modules, is

programming it to self-assemble into arbitrary 3D shapes. Our long-term goal is to use such

ensembles to achieve synthetic reality, an environment that, unlike virtual reality and augmented

reality, allows for the physical realization of all computer-generated objects. Hence, users will be

able to experience synthetic reality without any sensory augmentation, such as head-mounted

displays. They can also physically interact with any object in the system

in a natural way.

6

4. ClaytronicsThe enabling hardware technology behind synthetic reality is Claytronics, a form of

programmable matter that can organize itself into the shape of an object and render its outer

surface to match the visual appearance of that object. Claytronics is made up of individual

components, called catoms—for Claytronic atoms—that can move in three dimensions (in

relation to other catoms), adhere to other catoms to maintain a 3D shape, and compute state

information (with possible assistance from other catoms in the ensemble). In our preliminary

designs, each catom is a self-contained unit with a CPU, an energy store, a network device, a

video output device, one or more sensors, a means of locomotion, and a mechanism for adhering

to other catoms.

Creating a physical replica of an arbitrary moving 3D object that can be updated in real

time involves many challenges. The research involved in addressing these scientific challenges is

likely to have broad impact beyond synthetic reality. Particularly relevant to the ASPLOS

community, for example, is how to build and program a robust distributed system containing

millions of computers that must cooperate extensively in a harsh environment where their

configuration and goals are constantly changing? It is already possible to reproduce static 3D

objects. To create a dynamic 3D object, however, we will build upon ideas from such diverse

areas as modular and reconfigurable robotics and amorphous computing.

A Claytronics system forms a shape through the interaction of the individual catoms. For

example, suppose we wish to synthesize a physical “copy” of a person. The catoms would first

determine their relative location and orientation. Using that information they would then form a

network in a distributed fashion and organize themselves into a hierarchical structure, both to

improve locality and to facilitate the planning and coordination tasks. The goal(mimicking a

human form) would then be specified abstractly, perhaps as a series of “snapshots” or as a

collection of virtual deforming “forces”, and then broadcast to the catoms. Compilation of the

specification would then provide each catom with a local plan for achieving the desired global

shape. At this point, the catoms would start to move around each other using forces generated

on-board, either magnetically or electro statically, and adhere to each other using, for example, a

7

nano fiber-adhesive mechanism. Finally, the catoms on the surface would display an image;

rendering the color and texture characteristics of the source object. If the source object begins to

move, a concise description of the movements would be broadcast allowing the catoms to update

their positions by moving around each other. The end result is that the system appears to be a

single coordinated system.

8

5. Catoms The driving force behind programmable matter is the actual hardware that is

manipulating itself into whatever form is desired. Claytronics consists of a collection of

individual components called claytronic atoms, or catoms. In order to be viable, catoms need to

fit a set of criteria. First, catoms need to be able to move in three dimensions relative to each

other and be able to adhere to each other to form a three dimensional shape. Second, the catoms

need to be able to communicate with each other in an ensemble and be able to compute state

information, possibly with assistance from each other. Fundamentally, catoms consist of a CPU,

a network device for communication, a single pixel display, several sensors, an onboard battery,

and a means to adhere to one another.

Programmable matter consists of a collection of individual components, which we call

claytronic atoms or catoms. Catoms can:

•move in three dimensions in relation to other catoms

•adhere to other catoms to maintain microrobot; we believe that implementing a completely

autonomous microrobot is unnecessarily complex. Instead, we take a cue from cellular

reconfigurable robotics research to simplify the individual robot modules so that they are easier

to manufacture using high-volume methods.

At the macro-scale, catoms have a diameter > 1cm and weigh many tens of grams. In

light of the design principles stated above, the only viable force that can be used to move and

adhere catoms is magnetic; which sets a lower limit on the size and weight of a catom as the

magnets have considerable weight and volume. Furthermore, the circuitry needed for the high

currents necessary to switch the magnets increases the weight. At this scale, it may not be

possible to adhere to the “no static power” design principle. Our current prototype, as shown in

Figure 5.1, is a system composed of catoms that only operate in two dimensions. In this case

gravity holds the individual catoms to a surface and we do not have to deal with the adhesion

problem.

Much of the weight and size in a macro-scale catom comes from packaging, e.g., chip

packages, pins, wires, PCBs, etc.. In fact, in our planer prototype currently under construction we

9

estimate that more than 20% of the weight is packaging and more than 77% goes to the magnets

and their support circuits. This is partly because we are hand-assembling the catoms. Our next

version will use machine assembly which will allow us to shrink the packaging, resulting in an

estimated savings of twofold in weight and fivefold in volume.

Figure 5.1: The side and top views of a partially assembled planar prototype Catom.

Micro-scale catoms have diameters between 1mm and 1cm and weigh less than 1 gram.

Almost all packaging is eliminated and the catom is constructed by bonding VLSI dies directly

to MEMS-based sensor and actuation dies. The forces necessary to move a catom are now

sufficiently small that electrostatic forces become an option. Macro-scale catoms would require

electric field strengths in excess of the dielectric breakdown of air. Programmable nanofiber

adhesives (PNA), can be combined with electrostatic actuation to attach catoms without using

any static power. PNA is the next step in bio-mimetic adhesives which mainly use vander Waals

effects similar to that of the Gecko. The active components for micro-scale catoms are within the

realm of current engineering practices using micro- and nanoscale fabrication techniques. The

container could be manufactured by casting a spherical transparent polymer film and sputtering

indium-tin-oxide (ITO) electrode patches on the film. The patches of ITO would be used both to

connect catoms for distributing power as well as used to create capacitive coupling between

catoms in order to control their movement.

The next discontinuity occurs when catoms are manufactured using nanotechnology, e.g.,

as in chemically assembled electronic nanotechnology. In this regime, the catoms are small

10

enough, e.g., < 10 microns in diameter, that they are aerosol. While this is currently beyond the

state-of-the-art in manufacturing, further scaling of lithographic features in VLSI, and advances

in MEMS capabilities combined with advances in nanotechnology will enable the integrated

construction of such catoms.

In addition to capabilities, the different regimes (macro, micro, and nano) have

significantly different economics.

Macro-scale catoms require the assembly of multiple parts into a single unit. We expect

that this will make the realization of life-size synthetic reality prohibitive due to the cost per

catom. The micro-scale catoms may also require assembly, but with many fewer parts, e.g., no

boards and the elimination of magnets in favor of electrostatics. At this scale the entire catom

will be constructed using a parallel process, e.g., photolithography. Just as VLSI-based

computers are commonplace (as opposed to vacuum tube based computers), in this regime,

catoms are inexpensive enough that synthetic reality, though expensive, becomes viable. For

example, assume a 2mm catom controlled by 3mm 2 of silicon. If half the cost is in the silicon,

then, to construct a human sized form with an average depth of 10 catoms would require

approximately 1.25x10 7 catoms or less than one million dollars. In the nano regime they will

become sufficiently inexpensive that synthetic reality could be used not only for applications

such as telepario, but could be as ubiquitous as embedded processors are today.

11

6. Ensemble PrincipleRealizing this vision requires new ways of thinking about massive numbers of cooperating

millimeter-scale units. Most importantly, it demands simplifying and redesigning the software

and hardware used in each catom to reduce complexity and manufacturing cost and increase

robustness and reliability. For example, each catom must work cooperatively with others in the

ensemble to move, communicate, and obtain power. Consequently, our designs strictly adhere to

the ensemble principle: A robot module should include only enough functionality to contribute to

the ensemble’s desired functionality. Three early results of our research each highlight a key

aspect of the ensemble principle:

easy manufacturability

powering million-robot ensembles

surface contour control without global motion planning

6.1 High Volume manufacturability

Some catom designs will be easier to produce in mass quantity than others. Our present

exploration into the design space investigates modules without moving parts, which we see as an

intermediate stage to designing catoms suitable for high-volume manufacturing.

Figure 6.1 Claytronic atom prototypes. Each 44-mm-diameter catom is equipped with 24Electromagnets arranged in a pair of stacked rings.

In our present macroscale (44-mm diameter), cylindrical prototypes, shown in Figure 6.1,

each catom is equipped with 24 electromagnets arranged in a pair of stacked rings. To move, a

12

pair of catoms must first be in contact with another pair. Then, they must appropriately energize

the next set of magnets along each of their circumferences.

These prototypes demonstrate high speed reconfiguration—approximately 100 ms for a

step-reconfiguration involving uncoupling of two units, movement from one pair of contact

points to another, and recoupling at the next pair of contact points. At present, they don’t yet

incorporate any significant sensing or autonomy.

The current prototypes can only overcome the frictional forces opposing their own

horizontal movement, but downscaling will improve the force budget substantially. The resulting

force from two similarly energized magnet coils varies roughly with the inverse cube of distance,

whereas the flux due to a given coil varies with the square of the scale factor. Hence, the

potential force generated between two catoms varies linearly with scale. Meanwhile, mass varies

with the cube of scale.

6.2 Powering Microrobot Ensembles

Some energy requirements, such as effort to move versus gravity, scale with size. Others,

such as communication and computation, don’t. As micro robots (catoms) are scaled down, the

onboard battery’s weight and volume exceed those of the robots themselves.

To provide sufficient energy to each catom without incurring such a weight and volume

penalty, we’re developing methods for routing energy from an external source to all catoms in an

ensemble. For example, an ensemble could tap an environmental power source, such as a special

table with positive and negative electrodes, and route that power internally using catom-to catom

connections.

To simplify manufacturing and accelerate movement, we believe it’s necessary to avoid

using inter catom connectors that can carry both supply and ground via separate conductors

within the connector assembly. Such complex connectors can significantly increase

reconfiguration time.

Several members of the Claytronic team have recently developed power distribution

algorithms that satisfy these criteria. These algorithms require no knowledge of the ensemble

configuration lattice spacing, ensemble size, or shape or power-supply location. Further, they

require no on-catom power storage.

13

6.3 Shape Control Without Global Motion Planning

Classical approaches to creating an arbitrary shape from a group of modular robots

involve motion planning through high-dimensional search or gradient descent methods.

However, in the case of a million-robot ensemble, global search is unlikely to be tractable. Even

if a method could globally plan for the entire ensemble, the communications overhead required

to transmit individualized directions to each module would be very high. In addition, a global

plan would break down in the face of individual unit failure.

To address these concerns, we’re developing algorithms that can control shape without

requiring extensive planning or communication. While this work is just beginning, Claytronics

researchers have had early success using an approach inspired by semiconductor device physics.

This approach focuses on the motion of holes rather than that of robots per se. Given a uniform

hexagonal- packed plane of catoms, a hole is a circular void due to the absence of seven catoms.

Such a seven-catom hole can migrate through the ensemble by appropriate local motion of the

adjacent catoms.

Holes migrate through the ensemble as if moving on a frictionless plane, and bounce

back at the ensemble’s edges. Just as bouncing gas molecules exert pressure at the edges of a

balloon, bouncing holes interact frequently with each edge of the ensemble without the need for

global control. As Figure 6.2 illustrates, edges can contract by consuming a hole or expand by

creating a hole, purely under local control.

Figure 6.2 Hole motion. Edges can (a) contract by consuming a hole or (b) expand bycreating a hole, purely under local control.

14

We initiate shape formation by “filling” the ensemble with holes. Each hole receives an

independent, random velocity and begins to move around. A shape goal specifies the amount

each edge region must either contract or expand to match a desired target shape. A hole that hits

a contracting edge is consumed. In effect, the empty space that constitutes the hole moves to the

outside of the ensemble, pulling in the surface at that location. Similarly, expanding edges create

holes and inject them into the ensemble, pushing its contour out in the corresponding local

region.

Importantly, all edge contouring and hole motion can be accomplished using local rules,

and the overall shape of an ensemble can be programmed purely by communicating with the

catoms at the edges. Hence, we use probabilistic methods to achieve a deterministic result. Our

initial analyses of the corresponding 3D case suggest surface contour control will be possible via

a similar algorithm.

15

7. Software

The essence of claytronics a massively distributed system composed of numerous

resource-limited catoms raises significant software issues: specifying functionality, managing

concurrency, handling failure robustly, dealing with uncertain information, and controlling

resource usage. The software used to control claytronics must also scale to millions of catoms.

Thus, current software engineering practices, even as applied to distributed systems, may not be

suitable. We are just beginning to explore the software design principles needed.

We have broken down the software issues into three main categories: specification,

compilation, and runtime support. Our goal is to specify the global behavior of the system in a

direct and descriptive manner. The simplest model we are investigating with respect to

specification is what we call the Wood Sculpting model. In this model, a static goal shape is

specified. We are investigating two alternative compilation methodologies, both of which fit into

the general category of single-program-multiple data (SPMD) programming models. In the first,

we are compiling the specification into a planning problem. In this approach we are inspired by

work done in communicating soccer robots and in the context of reconfigurable robots, by the

constraint-based control framework in which a high-level description such as a particular gait

which is translated to a distributed, constraint-based controller. Our second approach is based on

emergent behavior. Prior work by seems particularly appropriate in this latter approach with

respect to claytronics. At the highest level of abstraction a shape is specified in terms of Origami

folding directives. Through a process of planning, these folding directives are translated into

low-level programs for autonomous agents; achieving the shape by local communication and

deformation only.

Underlying the user-level software is a distributed runtime system. This system needs to

shield the user from the details of using and managing the massive number of catoms. Our initial

steps in this direction use emergent behavior to determine a catoms location and orientation with

respect to all catoms as well as to build a hierarchical network for communication between

catoms. Efficient localization is achieved by having the catoms determine their relative location

and orientation in a distributed fashion. Then as regions of localized catoms join up they unify

their coordinate systems. Our algorithm takes O(1) time if the network is capable of broadcast.

With a network limited to point-to-point connections the algorithm takes O(√3 n) time in 3D.

Once catoms are localized we form a hierarchical communication network, again using simple

16

local programs on each catom. A tree is formed in parallel by having nodes join with their

neighbors until all the nodes are in a single tree. This simple algorithm produces a surprisingly

efficient tree from which can then be further optimized.

7.1 Meld: A Declarative Approach to Programming Ensembles

Meld is a programming language for modular robots, i.e., for independently executing

robots where inter-robot communication is limited to immediate neighbors. Meld is a declarative

language, based on P2, a logic programming language originally designed for programming

overlay networks. By using logic programming, the code for an ensemble of robots can be

written from a global perspective, as opposed to a large collection of independent robot views.

This greatly simplifies the thought process needed for programming large ensembles. Initial

experience shows that this also leads to a considerable reduction in code size and complexity.

Modular robotics is based on the concept of a (typically homogenous) collection of

robots working together to carry out a task. Most of the designs proposed to-date involve the

robots attaching together to form larger robotic structures. These connected ensembles gain

capabilities beyond those of the individual robots, including: greater strength from concerted,

parallel efforts; flexible, reconfigurable form to better deal with the environment and obstacles;

and greater capacity to redundantly sense the surroundings. However, planning and control for

the ensemble increases in complexity due to the large number of robots and exponentially large

configuration space. Compounding the problem is the fact that computation, sensing, and

actuation are distributed among the relatively limited robots—in general, no one robot is strong

enough to move all of the other robots, nor is it possible, typically, for any one robot to

determine the positions of every other robot or where they should be. These limitations in

strength and knowledge multiply the number of possible failure modes. For example, a robot

might move to the wrong location or physically impede the desired movement of other robots.

The ensemble as a whole might move into an unstable structure and risk collapse. To succeed,

the individual robots in a modular robotics system must work cooperatively.

17

8. Scaling and Design Principles

A fundamental requirement of Claytronics is that the system must scale to very large

numbers of interacting catoms. In addition to previously stated principles for the design of

modular robots we have the following four design principles:

Macro Micro Nano

Dimensions > 1 cm > 1 mm < 10 microns

Weight 10’s of grams 100’s of mg < 1 mg

Power < 2 Watts 10’s of mW 10’s of nW

Locomotive Mechanism

Programmable Magnets

Electrostatics Aerosol

Adhesive Mechanism Nanofiber Adhesives Programmable nanofiber adhesives

Molecular surface adhesion and covalent bonds

Manufacturing methods

Conventional manufacturingand assembly

Micro/Nano-fabricationand micro-assembly

Chemically directedself-assembly and fabrication

Resolution Low High HighCost $$$/catom $/catom Millicents/catom

Table 8.1: A summary of the characteristics of the different catom design regimes.

1. Each catom should be self-contained, in the sense of possessing everything necessary for

performing its own computation, communication, sensing, actuation, locomotion, and adhesion.

2. To support efficient routing of power and avoid excessive heat dissipation, no static power

should be required for adhesion after attachment.

3. The coordination of the catoms should be performed via local control. In particular, no

computation external to the ensemble should be necessary for individual catom execution.

4. For economic viability, manufacturability, and reliability, catoms should contain no moving

parts.

18

Catoms share many of the same features as individual robots in modular

(self )reconfigurable robots .In both claytronics and MRR the behavior of interest is group

behavior. The design goals of the individual robots and catoms are also similar in that the

individuals are self-contained. 1 The differences are however many. MRR are designed to

reconfigure their shapes in order to achieve functionality such as physical movement. While

scaling to many robots is a stated goal of MRR, it is not primary since typical goals (e.g., such as

moving across a terrain) require relatively few cooperating MRRs. In contrast, the original

motivating goal of claytronics is to create high-fidelity macro-scale 3D objects from micro-scale

components, as reflected in our design principles.

19

9. Process in Claytronics

As the size of the modules in a self-reconfiguring modular robotic system shrinks and the

number of modules increases, the flexibility of the system as a whole increases. In this paper, we

describe the manufacturing methods and mechanisms for a 1 millimeter diameter module which

can be mass produced. The module, which we call a catom, is the basic unit for claytronics, a

modular robotic system designed to scale to millions of units.

Figure 9.1 The Catom: Tube + Chip. Figure 9.2 Fabricated Catom.

Successfully realizing claytronics rests firmly on the ability to mass-produce sub-

millimeter scale catoms. Each catom must have a small robust physical structure integrated with

the ability to transfer energy to neighboring units, store energy, move, communicate with its

neighbors and selectively adhere to neighboring units; all under the local control. To invest each

unit with all of this functionality requires a novel approach towards the design and manufacture

of the catom.

Our design philosophy seeks to simplify the catom design by following the ensemble

principle, which states that: A unit should include only enough functionality to contribute to the

desired functionality of the ensemble. Adherence to the ensemble principle forces us to use local

computation and nearest-neighbor communication to simplify the physical mechanisms in each

catom as much as possible. In this instance, we use one physical mechanism (conductive plates

on, or near, the surface of the unit) for locomotion, adhesion, communication, and power

transfer. The local control of the catom has the ability to move charge between the plates. This

charge generates electrostatic forces between the robots to accomplish all the aforementioned

functions.

20

9.1 Introduction PathRelated Path

The design and implementation of a sub-millimeter autonomous robot, even one which

only needs to function in an ensemble, is challenging and therefore will be done over several

design generations. The first generation, detailed here, aims to develop the fabrication process

and implement the core functionality of the catoms: manufacturing, energy transfer, energy

storage, and actuation. Once this is achieved, functionality can be added to the catom through

relatively minor process improvements and adding additional circuitry.

Our approach, motivated by the need to mass produce the robots, is to integrate as many of the

individual components as possible. Our first design decision was to make each robot a sphere.

This simplifies manufacturing, while providing significant flexibility in the lattice arrangement

of the individual modules. A sphere can be created through stress-driven assembly of a photo

lithographically defined die. Locomotion is achieved by having the robots orbit each other using

electrostatic forces as described in these electrodes can be readily fabricated in two dimensions

before the device folds up. Having a solid approach towards realizing the structure, the next step

is to integrate active circuitry into the catoms. In order to simplify manufacturing and assembly

challenges, we decided to focus on integration and temporarily work on cylindrical tubes rather

than spheres. Using tubes relaxes the spatial constraints on interior components and allows us to

focus on improving the folding process to enable regions with changing radii, design of the

actuators, and integration of power transfer and active circuitry. Note that all of the techniques

developed are directly applicable to the final spherical shape so that simply decreasing the driver

circuit size and changing the initial two dimensional patterns will enable us to move from the

tube back to a spherical shape. The unit we present here represents all of the core capability

required in a catom so that moving forward is a matter of implementing new circuits and adding

functionality to the die.

Following the ensemble principle, we use conductive plates near the surface of the catom

to provide actuation. A first generation circuit capable of rectifying an incoming AC signal and

actuating the catom has been designed. Testing of the circuit has shown promise, but several

critical components require a redesign. These circuits are implemented using a commercial

CMOS foundry.

21

9.2 Manufacturing The Tube Micro robot

9.2.1 Fabrication of the Tube

The process for fabricating the catoms is critical to the goal of creating ensembles of

thousands to millions of them. Densely packed functionality must be provided through highly

repeatable low-cost mass production. Therefore, the process we are developing is based around

commercial CMOS processes. These processes are ideal for producing large numbers of

repeatable, compact, highly integrated circuits.In the long run, our fabrication process will have three distinct steps: CMOS foundry

fabrication, pick and place bonding, and three dimensional micromachining. The first two steps

in this process are both widely available from a number of sources, and therefore, our work to

date has focused on step three; a three dimensional micromachining process compatible with

prefabricated CMOS circuits and pick and place components. In developing this process, we

have chosen an approach based on stress driven self assembly of three dimensional structures.

Figure 9.3 The tubes are embedded in the wax Figure 9.4 The process steps for catom fabrication.(black) after the backing Si is etched.

The fabrication process used for the tubes is illustrated schematically in Fig. 9.4. It begins

(1) by thermally growing a 2μm thick layer of SiO2 on a silicon wafer. (2) The SiO2 is then

patterned using buffered oxide etch (BOE) to form rectangular islands that will later become the

tube microrobots. (3)Aluminum (Al0.99CuSi) of various thicknesses (0.1 μm to 0.5 μm) is

sputtered onto the SiO2 and patterned using a lift off process. At this point, the materials

required to form the tube are all in place and the next step is to free them from the original

silicon wafer. In order to do this, it is necessary to protect the aluminum, any CMOS circuits, or

22

mounted die from the silicon etch process. (4) Therefore, the wafer is mounted, device side

down, onto a sapphire handle wafer using Apiezon Wax W. This wax is very stable and is not

significantly attacked during the remaining process steps. Once the devices are mounted on the

handle wafer, (5) the back side of the silicon device wafer is etched. Since it is necessary to etch

over 0.5 mm of silicon, an initial etch is done using HNA (HF-Nitric-Acetic Acid). When the

device wafer thickness is below 100 μm, the rest of the etch is done using XeF2, an isotropic gas

phase silicon etchant that does not significantly etch most other CMOS circuit materials

including SiO2, Al, or photo resist. The final XeF2 etch stops cleanly on the SiO2 and wax

surfaces as shown in Fig. 4. (6) The devices are then released from the wax by immersion in

trichloroethylene (TCE).

Upon release of the tubes from the wax (7), the rectangular shaped tubes bend into

cylinders due to residual stresses in the SiO2 and Al. During the SiO2 thermal growth process,

oxygen atoms are introduced into a pre-defined silicon lattice SiO2. Not surprisingly, the

resulting oxide layer is highly stressed (≈ 350 MPa, compressive). In contrast, the sputtered Al

films can take on a wide range of stress values depending upon the deposition parameters. Under

most conditions, these films will have tensile stress with a lower magnitude (typically < 60 MPa,

tensile).

Upon release, the SiO2 expands more than the Al, causing the rectangular structure to

bend into a cylinder, with the Al on the inside. The physical shape of the oxide patch and the

pattern of the Al electrodes are used to control the direction of bending. In general, a rectangular

oxide patch will bend so that the long direction rolls up leaving the shorter axis as the length of

the tube.The radius of the tube depends heavily on the thickness of the layers. In general it is

possible to have different thicknesses or patterns of aluminum on the same tube, to create parts

with different radii. For example, in Fig. 9.3, the bottom six tubes have a solid aluminum patch

on the left side, which, when released, bends with a smaller radius of curvature. By controlling

the bending process in this way, it will be possible to mount a CMOS die on one end of the pre-

released structure. During folding the region the die is bonded to will stay flat, then a second

region starting at the edge of the chip will have a smaller radius of curvature, followed by an

outer region with a larger radius of curvature .Properly controlling the radius of curvature in

23

these two sections will enable us to keep the CMOS die close to the center of the tube. This is

crucial since a non-centered chip would make actuation more difficult.

Figure 9.5 A) After bonding the SOI CMOS chip to the tubes. B) After back-etching the chip. C) A released tube which requires re-rolling.

To further improve actuation performance and reduce manufacturing difficulty we etch

the back side silicon of the Silicon-on-insulator (SOI) CMOS die down to the SiO2 insulator

layer after it is bonded to the tube (See Fig 9.5). This substantially reduces the weight of the die

and makes the entire structure smoother, enhancing the release process.

9.2.2 Analysis of the Curvature

The radius of curvature for the released tubes can be predicted using the methods. In our

case, we have a two layer structure composed of SiO2 and Al. Each layer has a thickness ti,

Young’s modulus of Ei, a residual stress of σi. Fig. 6 shows calculated diameters for a range of

layer thicknesses assuming EAl = 70GPa, ESiO2 = 60GPa, σal = 60MPa, σSiO2 = −550MPa.

We measured thicknesses of tubes with both solid (Fig. 6, circles) and patterned Aluminum (Fig.

6, crosses). The patterned tubes have 100μm wide stripes of Aluminum electrodes with a pitch of

100μm, so the total aluminum coverage of the tube surface is 50%.

24

Figure 9.6 Tube diameter for different thicknesses of SiO2 and Al.

We see that for tSiO2 = 2μm and tAl = 0.5μm the expected diameter of a solid Al tube is

0.97mm. We choose to use tAl = 0.49μm and were able to achieve a diameter of 0.87mm. The

same thicknesses resulted in a 1.2mm tube when the aluminum is patterned. This increase in tube

diameter is expected, as with less aluminum coverage, there is less stress and thus less bending,

resulting in a larger diameter.

As shown in Fig. 9.6, the diameters of the tubes we manufactured are in agreement with

our analysis. Although thinner SiO2 gives a broader range of realizable diameters, the resulting

tubes are too fragile. Using a thickness of 2μm yields robust tubes with useful diameters.

9.3 Actuation

An important design goal in realizing claytronics is to use a single mechanism to perform

as many of the required functions as possible. The actuation mechanism for catoms is the most

challenging of the functions; it is essential that catoms are able to move and reconfigure in an

ensemble. In this section, we describe an actuation mechanism for catoms based on

electrostatics.

Catoms move by rolling along the surface of neighboring catoms. They use electrostatic

forces to do this. Each catom has conducting plates (“electrodes”) covering its outer shell; they

run along the length of the tube, e.g., in and out of the page in Fig. 7. The electrodes of

neighboring catoms align and form capacitors (see Fig. 9.7). When a voltage is applied between

25

the plates of a capacitor, attractive electrostatic forces are created due to the accumulated charge

on the electrodes. These forces then cause the catom to move.

Figure 9.7 The force created between the electrodes of adjacent catomsmoves the catom.

Since catoms are not electrically connected, they cannot share a common ground. In

order to generate a voltage across the electrodes, adjacent catoms are coupled using at least two

pairs of electrodes. This way, a series circuit loop is formed. The total electrostatic force created

between the electrodes causes the catoms to roll around each other until equilibrium is reached.

By switching the voltage on the electrodes, and successively charging/discharging the next pair,

a continuous rolling motion is possible.

9.3.1 Force Created Based On Voltage

Before we analyze how much force can be created, it is important to understand how

much force is needed to actuate a catom. For a catom to roll around a fixed catom, in the worst

case, the torque created by the electrostatic forces must be enough to overcome the moment

created by the weight of the catom, with respect to the point of contact. The moment created by

the weight of a catom is:

τrequired = MgR = 6.8 × 10−10Nm

where R is the radius of the catom, M is the total mass of the catom, including the chip.

26

Figure 9.8 The electrodes on neighboring catoms are paired to formcapacitors.

As shown in Fig. 9.8, the electrodes are placed radially on a catom, each spanning an

angle of θw. Adjacent electrodes on the same catom are separated by a gap, so that they do not

experience dielectric breakdown. In Fig. 9.8, when the catom on top is to roll clockwise around

the bottom catom, a voltage difference is applied between all the electrodes that are located in

the bottom right quadrant of the top catom and the electrodes that are located at the upper right

quadrant of the lower catom.

9.3.2 Required Voltage based on Eccentricity

Since no fabrication technique is perfect, is it reasonable to expect the catom to have a

non uniform shape after the fabrication steps explained. In this section we investigate the effect

of imperfect catom shape, and calculate the torque that is required in this case for a catom

moving horizontally.

We model our catom as a tube with an elliptical cross section, with the major axis a =

1mm, and a variable minor axis, b. We define the eccentricity of the catom as :

. Eccentricity is simply a measure of how distorted the catom is. With

our current fabrication technique as described, we are able to fabricate tubes that have minor-to-

major axis ratios ranging from 0.9 to 1, corresponding to an eccentricity of e = 0.43 to 0.

27

Figure 9.9 Ellipse with the maximum moment arm. For a : b = 4 : 3, the corresponding eccentricity is 0.66.

Using an analysis, a torque requirement is computed based on the eccentricity of catom.

We find the biggest moment arm of the center of mass of the catom, with respect to the point of

contact. Fig.9.9 shows this moment arm, k, which can be calculated analytically as

For the worst case measured eccentricity value of e = 0.43, the required voltage to

overcome the torque due to the imperfect shape is V = 18volts. This is well below the operation

voltage of the tube, V = 100V . Even if the chip size were to grow twenty fold, the required

voltage would still be manageable.We conclude that even in the worst case, it is possible to create a torque more than two

times needed to actuate the tubes made using our fabrication technique which yields tubes with

an eccentricity less than 0.43.

9.4 Realizing The Ensemble Principle

In the previous section we described the structures and methods for actuating catoms with

electrostatic forces. Here we briefly describe how these same structures can be used for adhesion,

communication, and power. The capacitively coupled electrodes of adjacent catoms create

electrostatic forces which are used for actuation and can clearly be used for adhesion, that is,

when the catoms come to equilibrium, the capacitively coupled electrodes hold them in place.

The use of capacitive coupling for inter-module latching was investigated in. Our analysis

showed that the bulk yield strength of an ensemble of catoms can be as high as that of plastics.

Power transfer is also possible using capacitively coupled electrodes. When one catom

puts an AC signal on its half of the plates of the capacitors, it generates an AC signal on the

adjacent plates of the neighboring catom. This signal can be rectified and converted to DC to

28

power the neighboring catom. In the worse case we obtain power efficiencies of 50%. In the best

case, when the catoms are trickle charging each other we can achieve efficiencies of nearly 80%.

Figure 9.10 Different radius tubes.

Finally, communication using the same electrode pairs is even simpler. Data modulated

with an AC signal can be easily transferred across the capacitors, and demodulated at the target.

9.5 Realizing The CatomThere are many steps involved in realizing a working catom. To this point we have made

progress in both manufacturing and actuation. The CMOS compatible fabrication process for

forming three dimensional tubes from a two dimensional process has been developed to the point

where tubes with two (or more) radii can consistently be fabricated at the same time, and even on

the same die. The actuation process has been modeled and simulated allowing the actuator

design to be optimized. Further, the simulations give us good confidence that the catoms can

provide sufficient force to operate once they are implemented, even accounting for non-

uniformities such as eccentricity. Proof of principle experiments have been done to demonstrate

rolling tubes using external actuation. Therefore, the next step is to integrate active control

circuits that collect and store energy and then use this energy to drive the catom.

Implementing the chip begins by selecting the appropriate CMOS foundry process. From the

actuation design work, it is clear that standard CMOS will not work as we may need voltages of

up to 100V to drive the electrodes. Further, we would like a process that uses Silicon-on-

insulator so that it can later be used as part of the shell forming process. Based on these

constraints, we selected XFAB’s XI10 High Voltage

BCDMOS process since it can support voltages in excess of 100V.

29

Figure 9.11 The chip.

Fig. 9.11 shows the schematic diagram of the circuits required to implement the first

catom. These circuits have all been to fit on a single die. The power required to operate the chip

is supplied through capacitive coupling. The tubes have electrodes running along the

circumference on both ends, called the “coupling electrodes” (See Fig. 16). An AC signal

generated at the source, i.e., a powered floor of another catom, is transferred to the chip through

the coupling electrodes. These AC signals are rectified to generate a DC output. The DC power

is then stored in a capacitor and used to power the integrated circuit and drive the electrodes at

voltages of up to 100v.

When completed, this first prototype catom will demonstrate the viability of a mass-

produced mm-scale robot with internal control and the use of a single mechanism—external

plates for capacitive coupling—which will be used for adhesion, actuation, and power transfer.

30

10. Applications

10.1 A 3D Fax Machine:This is a novel application of modular robotic technology. Many researchers expect

manufacturing technology will allow robot modules to be built at smaller and smaller scales, but

movement and actuation are increasingly difficult as dimensions shrink. We describe an

application—a 3D fax machine — which exploits inter-module communication and computation

without requiring self-reconfiguration. As a result, this application may be feasible sooner than

applications which depend upon modules being able to move themselves. In our new approach to

3D faxing, a large number of sub millimeter robot modules form intelligent “clay” which can be

reshaped via the external application of mechanical forces. This clay can act as a novel input

device, using intermodule localization techniques to acquire the shape of a 3D object by casting.

We describe software for such digital clay. We also describe how, when equipped with simple

inter-module latches, such clay can be used as a 3D output device. Finally, we evaluate results

from simulations which test how well our approach can replicate particular objects.

Consider a hypothetical scenario in the near future, where a paleontologist is searching in

a remote location for interesting fossil remains. Having found an interesting bone fragment, she

records the site information, using GPS and local surveying instruments, and takes numerous

digital photographs to record the context of the find. She then takes an impression of the

fragment in clay or plaster to facilitate its later reproduction for wider study as well as display in

museums. Given present technologies, the latency between discovery and access to fossil and

object replicas will generally be measured in months. However, imagine in this future world, that

the clay-like material used to take the impression is electronic in nature, comprised of a myriad

of tiny modules which are capable of inter-module communication and computation. This

intelligent clay measures its own shape and, by reflection, the shape of the embedded fossil

fragment, and generates a digital representation of the fossil’s three-dimensional structure.

31

11. Conclusions

11.1 Future Design

3Dmovement of catoms.

Catoms will be able to emit variable color and intensity of light, allowing for dynamic

physical rendering.

Onboard battery sufficient to power catoms will exceed the size of the catom itself, so an

alternate energy solution is desired

Research is being done into powering all of the catoms in an ensemble, utilizing the

catom-to- catom contact as a means of energy transport

Develop a unary connector for the catoms in order

Our initial research results suggest it may be possible to construct, power, and control large

microrobot ensembles to model 3D scenes. While many difficult problems remain, successful

implementation of a dynamic physical rendering system could open the door to a new era of

human computer interface design and new applications. Economic feasibility also poses a high

bar to the manufacture and deployment of multimillion-robot ensembles. However, the

innovative application of high- volume manufacturing techniques bridged a similarly large gulf

in cost and physical scaling in computer hardware. Achieving the Claytronics vision won’t be

straightforward or quick, but by taking on some of the problems associated with operating and

building these ensembles, we hope to advance the state of the art in modular reconfigurable

micro robotics and encourage others to undertake related research.

The advancements in nanotechnology and computing necessary for claytonics to become a

reality are feasible, but the challenges to overcome are daunting and will require great

innovation.

Many challenges need to be resolved before bringing this technology to real world.

The hardware aspect of this technology deals with making the crude and tiny robots to

move and configure.

The software aspect is still highly staggering.

It concerns with controlling the movements of minute catoms and programming these

millions to work together.

Still, claytronics is a promising field of engineering

32

As the capabilities of computing continue to develop and robotic modules shrink, claytronics

will become useful in many applications. The featured application of claytronics is a new mode

of communication. Claytronics will offer a more realistic sense to communication over long

distance called pario. Similar to how audio and video provide aural and visual stimulation, pario

provides an aural, visual and physical sensation. A user will be able to hear, see and touch the

one communicating with them in a realistic manner. Pario could be used effectively in many

professional disciplines from engineering design, education and healthcare to entertainment and

leisure activities such as video games.

33