COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub[let,jmc]" source
C00015 00003
C00017 ENDMK
C⊗;
.require "memo.pub[let,jmc]" source
.cb Chemical Memories Storing a Register per Molecules

	Avogadro's number, the number of molecules in a mole, is
6.023%8x%110%523%1.  If we could store a word or even a bit in a molecule,
the memory problem would be solved, so it seems worthwhile to ask whether
any fundamental laws of physics or chemistry prevent it.  The answer seems
to be that nothing prevents such memory densities, so it is just a
question of trying to invent a suitable scheme.  Here is one idea, but
there are lots more.

	A memory register is a chain molecule.  These molecules are
in solution or in a solid, but they are not in any fixed array,
so that it is not necessary to keep track of their locations unlike
the registers in present memories.  They are addressed by broadcasting
light or microwave signals throughout the container.  They answer by
emitting a photon which is picked up by a photomultiplier.

	Each radical on the chain represents a bit of the address
of the register.  Each has several bits of state some of may be
represented geometrically, i.e. the different states may be stereo
isomers of each other, and some temporary information is represented
by electronic excitation.  Transitions are brought about by signals
of particular wave lengths and are influenced by the state of the
radical and by the state of its neighbor to the "left".

	Thus each register is a kind of cellular automaton of a new sort.
The novelty is that a cell is affected by the states of its neigbors and
also by broadcast global signals.  Let us discuss first what kind of
cellular automaton we want before discussing how it might be realized
chemically.

Reading

	The register is a linear chain of cells, each storing one bit.  The
cells are identical except that the two end cells are special.  The
global signal has three values - 0, 1, and space.  A cell may %2contain%1
a 0 or a 1, and it may be in any of 3 excitation states,
i.e. it has six states.

	Cells are normally in excitation state 0.  A cell goes from
excitation state 0 to excitation state 1 if its left hand neigbor is
in excitation state 2 and the global signal matches its information bit.
It goes from excitation state 1 to excitation state 2 if the global
signal is a space.  In any other case it goes to or remains in excitation
state 0.

	A cell at the left end of a chain goes to excitation state 2
if an only if
it receives a special "start" signal.

	Now suppose that a signal consisting of "start" followed by
a sequence of 0's and 1's separated by spaces is given to a collection
of registers.  As long as the input sequence of 0's and 1's agrees
with the sequence of bits contained in the cells of the register, the
signal propagates along the register, but as soon as there is a discrepancy,
the signal dies.  Thus the excitation reaches the data region of the
register only if the address encoded in the global signal matches the
address of the register.

	The data part of a register is similar except that a cell can
be excited to fluoresce, i.e. emit a photon, provided its left neighbor
is in state 2, and the fluorescence exciter leaves the neighbor in state
2.  The wave length of the fluorescence depends on the information bit
of the cell.  There are two filters covering photon detectors - one that
admits only photons corresponding to a 0 data bit and one that only admits
photons corresponding to a 1 data bit.  Since the probability of detecting
the photon is not one, the fluorescence is excited repeatedly until
one of the photon detectors registers a hit.

	Now consider the chemistry of possible realizations of this
scheme.  We tentatively suppose that whether a cell %2contains%1 0 or 1
is represented by an isomeric state of the cell.  The 1 and 2 excitation
states are higher quantum states of parts of the cell.  The
output signal is the emission of a photon of an energy depending
on the bit stored.  The photons are detected by photomultipliers.
Since the detection is not perfectly efficient, we must either
excite the end cells repeatedly or have many registers with the same
information.  The former seems preferable.

	The big chemical problem is designing the cells so that their
excitability by the photons representing 0 and 1 depends appropriately
on their contents and on the excitation of their
left neighbors.  It should not depend on their right neighbors.

Writing

	With so much memory, it seems as though we should settle for
a "write once" memory.  Both the address and the contents are written.
Again the writing should require the left neighbor to be in an excited
state, so that the writing will propagate along the register.  Again
there should be a "start writing" signal, but it should be weak enough so
that it excites the left cells of only about the desired number of
unwritten registers.  A register that has previously been written should
have its start cell in a state that is immune to further "start writing"
signals.  It would be best if the writing could be done in a memory
box that contained written registers, but if necessary, we could imagine
that registers are written separately and then added to the memory box.
Since it would involve physical motion, it should be avoided if possible.

	We want the 0's and 1's in the memory to be very stable
so that they don't make spontaneous transitions, and this suggests
stereo isomers.  The unwritten state should be a third state of the
cell that can make a transition to a 0 or 1 state, but requires for
this transition that the left neighbor be already written.

	Except for the final fluorescence which takes a few nanoseconds,
each of the other steps can apparently be accomplished in a few
picoseconds if it is desired to read and write that fast.

	I don't understand the chemical problems of realizing this
kind of memory cell, but if this doesn't work, other schemes can
be devised.

	If it is difficult to design cells with six states, we can get
by with cells having only two states as follows.  There are three
kinds of cells: data cells and left and right excitation cells.  If
a data cell's left neighbor (a left excitation cell) is excited and
the global signal matches its data bit, then the cell on its right
(a right excitation cell) is excited.  A space passes the excitation
from a right excitation cell to the left excitation cell to the right
of it.  It may be feasible to use the same kind of cell for left and
right excitation cells.

	If the pulse length of the global signal is
short enough (a very few picoseconds), it may be possible to dispense
with the space signal.  Its purpose is to prevent an addressing signal
from propagating down several address bits that are the same, as might
happen if a data cell excites its neighbor directly.


	Here is a variant that may be easier to realize.  There are three
kinds of radicals in the chain called A, B and C, and they alternate
cyclically, i.e. the chain has the form ABCABCABC....  Radicals of type
A have two excitation states 0 and 1.  Radicals of type B have two data
states 0 and 1, and radicals of type C have two excitation states.
Most of the time the radicals of types A and C are in their
0 states, and when this is the case, they are unresponsive to stimulation.
However, when an A is in state 1, it is sensitive to broadcast input
signals 0 and 1, becoming instantaneously excited and emitting signals
of types 0 and 1 to its neighbors.  A neighbor of type C is insensitive
to this stimulation, but a neighbor of type B will relay a signal provided
the signal it receives from A (which corresponds to the broadcast bit)
agrees with its state.  This relayed signal will excite the adjacent
radical of type C into its state 1.  A broadcast signal "space" propagates
the excitation from the radical of type C to the adjacent A.  The
purpose of having type C is analogous to the use of a three phase clock
in circuits made of diodes, i.e. to make sure the signals propagate only
in the desired direction.

	The following scheme may be easier to implement.  Instead of
8 states, a cell has just 4 states.  One bit is its information bit
and the other is an excitation bit.  However, the information cells
are interspersed with punctuation cells.  Punctuation cells have only
excitation bits, and a punctuation cell is excited when its left
neighbor is excited and it receives the punctuation-excitation
frequency.  An information cell is excited when the punctuation
cell on its left is excited and the broadcast information-excitation
signal matches its information bit.  The advantage of this scheme
is that the individual cells can be less complicated.

.nofill
.turn on "∂"

∂(40)John McCarthy

This draft of CHEMEM[W79,JMC] pubbed at {time} on {date}.