In practice, electronic computers require huge amount of energy and they generate a lot of heat while computing. As a result the task of cooling down the computer components is a central problem in computer engineering. This problem prompted several prominent computer scientists such as C. H. Bennett, R. Landauer, E. Fredkin and T. Toffoli to investigate intrinsic limits on a computer energy consumption and generation of waste heat. Their research, spanning three decades, came to a conclusion that a computation process may be performed reversibly, generating zero waste. Moreover, they concluded that most computer operations could be performed without external energy, although such computation would take infinitely long time and introduce high error rate. However, one operation was singled out for its intrinsic demand for energy investment. This is the Îresetâ operation: setting an unknown bit to zero. The Îresetâ operation would decrease the entropy of the system, which must be compensated by putting some energy into the system in order to comply to the second law of thermodynamics.

ÎEnergy storageâ is somewhat evasive term. From high-school mechanics we learn that potential energy is something that may be exploited to do some work. A physical system tends to rearrange from a state with high energy to a state with lower energy and perform some work due course. A stone raised above the ground gains potential energy and as it falls, it may break whatever it hits, or just heat the ground a little bit. So the notion of potential or stored energy is not an absolute but rather a relative property. In a same sense, a DNA polymer molecule stores significant amount of energy relative to the same polymer broke into little pieces - its monomers. The DNA molecules do not fall instantly apart due to high energetic barriers that confine the polymer in a kind of an energetic Îwellâ. However, nuclease enzymes take the polymer over these energetic barriers all the way to breaking apart. As a result, energy is released. The fate of this energy may depend on whether we may utilize it or not. For example, breaking of the chemical bonds of the ATP, which is chemically similar to the kind of bond broken in DNA decomposition, is exploited by the cell to fuel numerous biomolecular transformations. This is in essence equivalent to throwing down a stone in order to raise some other stone further up using a lever. Energy released by DNA decomposition may also be exploited to do useful things, such as molecular computation.
Another kind of energy may be stored in single-stranded DNA molecules relative to the hybridized, double-stranded version. In this process no chemical bonds are either broken or formed, but rather many weak non-covalent bonds are formed.

Yes. A number of nano-scale DNA-based mechanical devices were designed by B. Yurke and colleagues, and N. Seeman and colleagues. These devices perform conformational changes as a result of various hybridization processes between their single-stranded DNA components and external single-stranded DNA serving as Îfuelâ. As far as we know, decomposition of DNA was never used to drive any useful processes.
 

As we mentioned, DNA would break into pieces unless high energetic barriers would prevent it from doing it. Nuclease enzymes lower these barriers and allow DNA to decompose. Some of the nucleases cleave very precisely at certain locations, determined by the exact sequence of the nucleotides at or near their cleavage sites. Yet, even such specific cleavage would only generate useless heat. In our molecular computer we use this energy to drive a useful process, namely computation. It should be understood that on the physico-chemical level we have a sequence of chemical reactions, resulting in a certain chemical product. This chemical process would not happen if energy would not be released at each step. The energy is indeed released because we break DNA into smaller pieces with the help of FokI restriction nuclease. At a conceptual level, the sequence of chemical reactions are steps of computation, and the final product is a computational output. The whole computation process and output formation depend on the energy stored in the starting material, a molecule encoding an input data. Therefore, the computation (an abstract notion) is made possible to the (physical) energy stored in the (physical implementation of its) input relative to the energy stored in its output. There are no other sources of energy; the computation is fueled solely by its input. 

Generally, a state machine consists of 1) a data tape divided into cells, each containing a symbol selected from the tape alphabet, and 2) a computing device driven by transition rules. The device is positioned over one of the cells and is in one of a finite number of internal states. Depending on the symbol detected and on the internal state, a transition rule instructs the device to write a new symbol, change state and move one cell to the left or to the right. The Turing machine is the most general type of state machine, capable of writing on the tape as well as moving in both directions. 
A more restricted, yet important, class of state machines is finite automata. A finite automaton is a unidirectional read-only Turing machine. Its input is a finite string of symbols. It is initially positioned on the leftmost input symbol in a default initial state, and in each transition moves one symbol to the right, possibly changing its internal state. Its software consists of transition rules, each specifying a next state based on the current state and current symbol. A computation terminates after the last input symbol is processed, the final state being its ãoutputä. Alternatively, it may suspend without completion when no transition rule applies. Some states are deemed accepting. An automaton accepts an input if there is a computation with this input that ends in an accepting final state. A finite automaton with two states and an alphabet of two symbols is shown in Fig. 1A. It determines if an input over the alphabet {a, b} contains an even number of aâs. Fig. 1B shows the intermediate configurations obtained during a computation of this automaton. A two-state two-symbol automaton can have eight possible transition rules. Programming such an automaton amounts to selecting transition rules and deciding which states are accepting. A selection of such programs is shown in Fig. 1C
.

 

Fig. 1. Finite automata and inputs for which we show molecular realizations. (A) Automaton A1. Incoming unlabeled arrow marks the initial state, and labeled arrows represent transition rules. A double circle represents an accepting state. (B) A computation of A1 scanning and digesting the input abba. Each row represents an intermediate configuration showing current state and remaining input. The transition rule applied is shown in brackets. (C) The two other automata for which we demonstrate molecular operation.
 

Similarities:
1) Both implement two-state two-symbol finite automata
2) Both have stretches of double-stranded DNA to encode the input strings. Both work on artificial inputs of well-defined structure rather than on any DNA molecule
3) Both encode current state and symbol in a single-stranded sub-sequence of the input symbols.

Differences:
1) The new computer does not use ATP and is fueled by its input; the old one relied on ATP as its energy source.
2) The new computer recycles its software and hardware; the old one consumed one software molecule per transition.
3) The new computer is 50 times faster and performs 8000 times more operations in parallel than the old one
4) The new computer can process much longer input strings. We could routinely compute on inputs 12 symbols long but also got results with 20-symbols long strings. The yield of a single step rose above 98%. The old one could only process 6-symbol inputs on a good day, with about 50% single step yield
5) The new computer has a completely redesigned structure, making possible all the improvements summarized above
 

No. The error rate is more or less constant under different conditions employed

All DNA molecules are ordered from a manufacturer that supplies single-stranded DNA molecules according to specification (exact sequence, amount, mode of purification). We then purify the molecules and employ various chemical procedures to get to the desired molecular structures (such as stepwise ligation of DNA blocks to build long synthetic inputs)

We mix all the components of the computer: its radioactively-labeled input, software and hardware in 10 microliter of solution and let the process run for several minutes at 8^oC. We then analyze the outcome by denaturing gel.
 

Using naturally occurring enzyme and radioactive ATP molecules we can put a radioactive isotope of phosphorous on a terminus of a DNA strand. This label is very sensitive and can supply quantitative information.

You have a gel with two sides attached to electrodes, and place the DNA near the minus electrode. The gel is heated and it contains urea, which causes strand dissociation, so that each single strand is on its own. The DNA strands travel slowly inside the gel towards the plus electrode, at a speed that is a function of the molecule size. If you stop the process at the right time, you have a spread of the molecules according to their length. If you add on the side a DNA mix with a "ladder" of DNA molecules of known lengths, you can determine the lengths of the different DNA molecules in the original solution. Some of the molecules are labeled by radioactive phosphorous, and we can see their location on a gel using special sensitive plates and readout device.

In a feat of molecular nanotechnology, scientists at the Weizmann Institute were able to construct a self-contained, programmable computing device using only three types of molecules --- a DNA input molecule, encoding the data and providing the fuel for the computation, DNA software molecules, encoding the rules of computation, and a hardware molecule, a DNA-cutting enzyme. Combining an input data and an energy supply for the computation in the same physical entity is unthinkable of in the realm of electronic computers. The computing device is so small that 3 trillion such devices can reside in one microliter of solution [or - 60 trillion devices in a teardrop], a fact that won it the Guinness World Record for the world's smallest biologic al computing device.

All electronic computers work according to a mathematical model developed by John von Neumann in the 1940s, called the stored program computer. In it both program and data are stored as words in the computer memory. Each memory word can be accessed by its address. The electronic computer fetches program instructions one by one from its memory and executes them. Typically the instructions are to load data from a certain memory location into a register, to store the content of a register into a memory location, or to perform an operation on registers, for example to add the content of two registers and store the result in a third register.

The stored program computer model is ideal for realization using electronic circuits, and hence it has been the basis of all electronic computers for more than 50 years.

Our molecular computer is based on a very different model, called a finite automaton, which is a simplified version of the Turing machine, a mathematical model of computation developed by the British mathematician Alan Turing in the 1930s. It has data stored as a string of symbols (Turing envisioned a paper tape divided into squares, each holding one symbol) and a control unit, which holds the program rules, is in a particular location on the string and can be in one of a finite number of states. The Turing machine starts the computation in an initiate state and on the leftmost symbol in then string. In each cycle a Turing machine reads a symbol and executes a program rule that specifies what action to take according to the symbol read and the internal state. The actions can be to replaces the symbol with a new symbol, move one symbol to the left or to the right, and/or change internal state. A finite automaton is a Turing machine that cannot change the string of symbols, and in each step moves one symbol to the right. Hence the result of the computation of the finite automaton is the final state reached upon reaching the last input symbol.

Both the stored program computer and the Turing machine are so-called "universal computers" and as such have equivalent computing power. A finite automaton is much weaker.

Our molecular finite automaton has three components. An input molecule, which stores the data to be processed as well as the fuel needed for the computation, software molecules, which encode program rules, and a hardware molecule which performs the computation, as directed by the software and the input. The molecules float in a watery solution containing some salts needed for the operation of the hardware molecule.

The input molecule is a double-stranded DNA molecule. A mathematical trick is used to store the current state and current symbol to be processed in a "sticky-end" of the DNA molecule, a single-stranded protrusion obtained by having one strand longer than the other. The software molecules also have "sticky-ends", each matching a different state-and-symbol input combination. Through a spontaneous chemical process called DNA hybridization, the input molecule and a matching software molecule temporarily connect to each other. The hardware is a naturally occurring molecule, an enzyme called FokI that attaches to DNA in a specifically coded location and cuts both strands at a fixed distance from that location, creating a sticky end. Using additional chemical and mathematical tricks, the software molecules are designed so that the hardware enzyme attaches to them and, once a software molecule hybridizes with an input molecule, the hardware molecule cuts the input molecule in a programmed location, exposing a new sticky end that encodes the next state and the next symbol to be processed. The final sticky end, obtained at the end of the computation, encodes the computation result.
 

Our molecular computing device realizes a so-called "two-state two-symbol finite automaton", which can do a rather limited set of computations, depending on which program molecules are used (i.e. mixed in the watery solution). For example it can detect if a list of 0's and 1's has an even number of 1's; if it has at least one 0; if it has no two consecutive 1's; if it has no 0's after a 1; or if it starts with 1 and ends in 0.
 

Each molecular computer is rather slow, on the average it performs one operation per 20 seconds. Also, a PC is a "universal computer" whereas out molecular finite automaton is very limited in its computing capabilities. Nevertheless, if we consider the size and energy consumption, then a spoonful (5 milliliters) of our "computer soup" contains 15,000 trillion computers that together perform 330 trillion operations per second, which is more than 100,000 times faster than the fastest PC. This spoonful releases less than 25 millionth of a Watt, hence its energy-efficiency is more than a million times that of a PC.
 

This is basic research at an early stage, it may be too early to predict its impact. However we envision biomolecular computing devices operating as "smart components" in biochemical reactions, which may have pharmaceutical and biomedical applications in the future.
 

In the normal course of events it may take decades for this research to be applied. But basic research sometimes has surprising and unexpected ways in which it may be applicable.
 

The key conceptual advance, compared to our earlier molecular automaton published more than a year ago, is that the new automaton uses its DNA input as its source of fuel, hence it does not need an external fuel supply. In addition, in our previous molecular automaton, each computational step consumed one software molecule, and hence the computation needed an ongoing supply of software molecules. The new automaton does not, so the net result is that a fixed set of software and hardware molecules can process any input molecule, of any length, without an external supply of energy.
 

The field started in 1994 with the publication of a paper by Len Adleman, describing how one can use DNA to solve combinatorial problems, such as the so-called "traveling salesman problem" which is to find an optimal route through several cities, given distances between them. The hope was that the parallel nature of DNA manipulation can be used to outperform electronic computers in solving such problems.

This approach involves multiple outside interventions during computation. A solution containing the molecular components is subjected to mixing, separations, PCR amplification, etc several times during computation. Therefore, it is not truly "molecular computers" but rather "lab-assisted molecular computers", where the lab personnel plays a crucial role in accomplishing a computation.

 Today fewer people believe this can be achieved due to inherent limitations of the approach and the rapid advance of electronic computers. Our work does not attempt to compete with electronic computers head-on, but rather to design molecular-scale computing devices, which might have applications in areas not accessible to electronic computers, such as "smart components" in biochemical reactions.

Another line of research is "semi-autonomous molecular computation", which is closer to our work. This direction is pursued by Hagiya and Sakamoto in Tokyo University, and by Winfree, Reif and Seeman in the USA. This approach aims at creating systems that evolve in a programmed fashion to perform a computation, while minimizing the external intervention. These researches were successful in performing very simple computations without much interference except regulating the temperature during the process (in a PCR manner - multiple cycles of heating/cooling). Still, experimentally demonstrated computation capabilities of these systems is substantially lower than that of our devices, and they are not completely autonomous.

The self-sufficiency is crucial for the future success of such devices, as we cannot control the processes on the cellular level from the outside. Moreover, our recent result demonstrates that useful computation may be performed on the expense of biopolymer degradation. As biopolymers are constantly degraded in a cell, we might envision "smart components" that utilize this degradation without placing  any additional energetic burden on a cell.
 

No, we are aiming at molecular computers that operate in a biochemical environment.
 

We synthesized the molecule to containing the input information we wanted to test in a computation.  For example if we wanted to run the computation on the list "aabba" we synthesized a DNA molecule that encodes that list, using standard lab techniques.  The energy is inherent to the chemical structure of DNA, and is independent of the actual data it encodes, only on the length of the molecule.
 

DNA synthesized in a lab using standard techniques, we purchased it according to out spec from outside supplier. FokI enzyme originally comes from a bacteria named Flavobacterium okeanokoites. It is currently produced with the help of recombinant E. coli strain that overexpresses the gene for this enzyme.
 

Yes, all these chemical reactions can only happen in water solution at certain pH values and salt concentrations. These conditions (pH and salt) are required to allow the normal function of the hardware enzyme FokI.
 

Theoretically no, but enzymes are known to get "tired" after a while, so fresh enzyme supply, and cleaning debris, might be needed after long computations.
 

"Computer" normally refers to general-purpose programmable universal computing device.  Our computer is programmable but is not universal (there are computing tasks it inherently can't do) since it is a realization of a weak model of computation called "finite automaton".  Hence the caution in the phrasing.
 

We use 8 base pairs per input symbol, with some tail, for the input, and short oligos for the software molecules, with one strand 4 nt longer than the other (Fig. 2 of the PNAS paper). Each of the software molecules is about 20 base pairs long. They are artificially constructed molecules, much much shorter than human DNA.
 

The old molecular automaton did not work without ATP, although it could make one or two steps, encouraging us to investigate the phenomenon more closely. Making FokI to cleave DNA via noncovalent bridge proved to be a tricky task and required many trial-and-error attempts before arriving at a working design. Making the computer work fast was another problem and we had to use a hardware enzyme in a stochiometric rather than catalytic amount.