91av

WHEN Jim Erickson went to bed around midnight on 9 October 1999, his mind was
half a billion kilometres away. And his fingers were crossed. He was thinking
about Galileo—NASA’s gold and black spacecraft hurtling towards its
rendezvous with Jupiter’s moon Io—worrying that it was ploughing through
an intense belt of radiation that could grill its circuits and turn it into a
lump of useless junk.

At the time, Erickson was Galileo’s project manager at NASA’s Jet Propulsion
Laboratory in Pasadena, California, so when his phone rang at 3 am he feared the
worst. “There is this immediate feeling in the pit of your stomach that, oh
Jeez! this is the wrong time for something to happen,” says Erickson.

Radiation, it turned out, had zapped a critical bit of Galileo’s computer
memory and the spacecraft had partly shut itself down to prevent further
disaster. Erickson and his engineers at JPL’s mission control raced to
redesign the software and uploaded it to Galileo just in time to watch the craft
soar successfully past Io.

Engineers like Erickson know only too well that spacecraft electronics are
extremely susceptible to the high levels of radiation and violent swings of
temperature common in space. Often there’s little they can do once their
precious craft is launched—except pray.

Now, however, scientists have begun to build electronics that will one day
relish such extreme environments, and make Darwin proud into the bargain. When
pushed to the brink of extinction, their designs mutate, compete and eventually
evolve into circuits that are fighting fit once more. Other scientists are
mimicking some of life’s essential structures—cells, genes and even whole
organisms—to create hardware that replicates and heals itself. The result
seems less like electronics and more like an emerging species ideally suited for
exploring the deep sea or the harshest regions of the Solar System.

The earliest ancestors of this species emerged from the electronic clutter on
the lab bench in the mid-1990s. Researchers were starting to experiment with
field-programmable gate arrays (FPGAs), which consist of an array of logic
gates—each built from dozens of transistors—interconnected by
switches. Since engineers can choose which switches are open or closed, they can
control how the logic gates in the circuit are connected, reconfiguring the
system at will to make all kinds of simple electronics.

As software determines which switches are on and which off, engineers can
even hand over control to a special kind of program called a “genetic
algorithm”. This encodes the state of the switches as a stream of 1s and
0s—representing open or closed switches respectively. It then uses these
strings of binary data as “genetic material”, adding mutations or combining
different strings to create a range of new circuits in much the same way that
living creatures evolve by shuffling DNA. Then the circuits are tested against a
desired result and the “genes” of the best ones are fed back into the genetic
algorithm so evolution can continue.

Researchers have already discovered that they can dramatically speed up
circuit design by combining FPGAs and genetic algorithms, and even create
electronics with less than a tenth of the components that a human designer would
have used (91av, 15 November 1997, p 30).

However, FPGAs are not suitable for designing spacecraft electronics as they
can only be used to build digital circuits and not analogue circuits of the type
found in sensors needed for space exploration. “Analogue circuitry is a vital
component of spacecraft electronics,” says Adrian Stoica, an electrical engineer
at JPL. Analogue electronics is vital for infrared sensing and high-frequency
communications, for example. Processing information using analogue circuits
before converting the information into digital data increases efficiency and
analogue systems generally consume less power than their digital counterparts,
says Stoica.

So in 1998 Stoica embarked on a project to create a more versatile variant of
the FPGA that he calls the field-programmable transistor array. His FPTA is a
multicellular system, where each cell is a chip containing 24 programmable
switches and an array of 8 transistors, rather than the dozens of transistors
found in an FPGA. When their switches are programmed, FPTAs make the perfect
building blocks for all kinds of analogue or digital circuits.

Stoica thinks these circuits will even be able to recover from faults
triggered by biting-cold temperatures or the huge doses of radiation that
spacecraft such as the Galileo orbiter or Cassini probe encounter as they pass
Jupiter. Energy from ionising radiation, for example, can create spurious
electrical signals that damage transistors or alter the settings of switches.
Build evolutionary software into future spacecraft and it should be able to heal
broken circuits on the fly.

Stoica and his colleagues Didier Keymeulen and Ricardo Salem Zebulum began by
using a genetic algorithm to design a circuit for a particular job, namely an
analogue filter—like those used to select channels on a television or
radio—for use in a communications module. After the algorithm had created
a circuit that performed satisfactorily, they tested it under conditions it
might encounter in space. First they plunged the circuit into a vat of bubbling
liquid nitrogen at –196 °C. Then they roasted it in an oven at 250
°C. Finally the team induced faults in the circuit by simply disconnecting
some of the connections between the chips. The big question remained: could
evolution repair the damage?

Fast forward

The results were impressive. Once the computer controlling the filter had
detected that the circuit was not functioning properly, it restarted the genetic
algorithm to search for another circuit that could do the same job, using a
modified arrangement of transistors and switches that worked around the fault.
They quickly discovered that the new evolutionary cycle took half the time that
the algorithm had taken to design the original circuit. This makes sense,
because the computer already had a copy of the last generation of individuals
that had been evolved, and the genetic algorithm simply used these individuals
as the starting point for a new evolutionary cycle. Even nature, in the event of
a catastrophe, doesn’t go back to the primordial soup and evolve bacteria from
scratch—it starts with what life is available at any point in time.

The implications for space electronics are tremendous. If faults occur when
the system is billions of kilometres from Earth, evolutionary software could
quickly repair the circuits without the kind of frenetic activity back on Earth
that Galileo needed—or worse, total mission failure. But while Stoica
believes that evolutionary hardware is perfect for sensor electronics in
spacecraft, he admits that for the moment it shouldn’t be used to build critical
systems such as the thruster controls or anything that might accidentally send
the craft careering off in the wrong direction while the electronics is
repairing itself.

And what if the computers that control the evolutionary process were damaged?
The solution, Stoica believes, lies in a “die-hard” architecture where
everything—the hardware, electronics and computer controllers—is
built using a self-repairing, self-reconfiguring architecture. “Maybe more like
an array of cells in which each has some reconfiguration function and some
recovery,” he says. “It would have a homogeneous structure where it doesn’t
matter where the fault is, it will recover.”

This dream is closer to reality than you might think. On the other side of
the Atlantic, a team of researchers at the Swiss Federal Institute of Technology
in Lausanne is busy developing a complete hardware system that will not die.
Last year, working in labs near the shores of Lake Geneva, they finally
completed a remarkable prototype machine that they call the “bio-watch”.

It comprises three pairs of modules, each pair displaying hours, minutes and
seconds. On each module is a kill button. Pressing it destroys the module, yet
if you kill one a spare module takes its place. The watch, like a living
creature, immediately repairs itself.

Daniel Mange, Moshe Sipper and their colleagues at the institute have named
their project “embryonic electronics”, or embryonics, and their goal is to
create an artificial electronic organism, replete with molecules, cells, genes
and a genome capable of surviving anything that nature can throw at it.

Silicon creatures

The resilience of their embryonic circuits comes from the way they are
constructed. Much like living things, the basic building blocks of embryonics
are elements that the team calls cells
(see Diagram). Each cell contains
hundreds of “molecules”, each made of multiplexers—small groups of
transistors much like the FPTAs in Stoica’s circuits. “It’s possible to show
that such an element is completely universal,” says Mange. “With enough of these
elements you can construct any logic function.”

Just as animals and plants rely on their genes to control their cells, an
embryonic circuit relies on a software “genetic code” to organise its molecules
and cells. Mange and his team begin by constructing an array of thousands of
molecules. Their next job is to convert these molecules into an “organism” made
up of cells. To do this, they download a “polymerase genome” program, named
after the polymerase enzyme found in living cells. In the case of the bio-watch,
the polymerase genome contains instructions required to group the molecules into
cells of approximately 600 molecules each.FIG-mg22763801.JPG

With the basic size of the cell defined, the researchers then download a
second program called the ribosomic genome. This configures the switches in each
cell so that its molecules form a processor, complete with its own memory for
data storage. Each cell is now capable of executing programs exactly like a
processor chip in a computer.

Finally, they load software that organises these cells so they work together
to make a functioning piece of hardware—or an organism—such as the
bio-watch. This software, called the operative genome, is injected into the cell
at one corner of the organism, called the mother cell. The mother cell then
passes the genome to two adjacent cells, which pass the genome on to their
neighbours until the whole organism is programmed.

The operative genome contains specific instructions for each cell in the
organism, so that together the cells can form a working system. This process is
analogous to cell differentiation, which is responsible for turning a mass of
identical embryonic stem cells into specialised cells such as nerve and
muscle.

Before the cells run the operative genome, each calculates its coordinates in
relation to the mother cell. Then each cell executes the portion of the
operative genome that is specific to its coordinates. In the bio-watch, for
example, a pair of cells calculates seconds, another pair minutes and a third
hours. The same genome is operating in each cell, but the cells play different
roles—just like specialised cells in a living organism.

One of the biggest advantages of this approach is that simply by loading a
different set of polymerase, ribosomic and operative genomes, the hardware can
quickly reconfigure itself into an organism suited to a completely different
task. This could be especially useful in space exploration. Imagine a robot
trundling across the surface of a distant planet. While it searches for signs of
life with its cameras, its embryonic computer would be configured for image
processing. When it decides to send its findings back to Earth, it could load up
a new set of genomes which adjust its processor for data compression and
communication. A robot carrying multiple genomes would be capable of
reconfiguring itself for a variety of other tasks and missions, from swimming in
alien oceans to setting up a chemical factory to manufacturing oxygen or rocket
fuel.

The other benefit of mimicking nature is that the modular approach is ideal
for self-repair. Inspired by DNA’s self-repairing double helix, Mange and his
team have doubled up every molecule in a cell and made these pairs perform as
one. When both work perfectly, everything’s fine. But if one of a pair
malfunctions, its partner automatically triggers a fault signal. The cell then
switches off the faulty pair and reconfigures the circuit around it using
molecules that the cell holds in reserve.

If a large proportion of molecules in a cell are faulty, the whole cell will
switch off. This automatically triggers reconfiguration at the level of the
organism. If spare cells are available, the organism changes the connections
around the rotten cell. Given enough spares, it can even replicate an entirely
new copy of itself.

For now, Mange and his team are testing embryonics on simple systems. In
addition to the bio-watch, they have created a stopwatch and a random number
generator. They are also embarking on a challenge familiar to all computer
scientists—they want to build a universal Turing machine. This device
would show whether any mathematical theory is true or not.

Stoica, too, is interested in the embryonics approach. “We think it has a lot
of potential for our applications,” he says. Stoica’s project is itself ready to
take the next step—evolving more complex systems using algorithms and
integrated circuits with 1024 transistors and then turning up the heat in the
ovens to 600 °C.

Recycled chips

Embryonics may even provide unexpected bounty closer to home. As the circuits
on processor chips shrink, tolerances in semiconductor fabrication plants get
ever tighter. This means that the manufacturers are forced to throw away nearly
90 per cent of their chips due to faults. “Let’s take the 90 per cent and make
use of them, despite the faults,” says Sipper. “And how do you do that?
Self-repair.” Sipper believes these rejects could be used to build embryonic
circuits that would compensate for the faults on individual chips.

Self-repairing hardware could also be a life-saver for processors working
inside radiation-riddled nuclear reactors or in robotic systems for deep-sea
exploration, where human intervention would be all but impossible in the event
of a disaster, says Sipper.

With an artificial organism that has a genome, would it be long before such a
machine began to evolve spontaneously? That’s our main goal, admits Mange.
“Evolving an artificial genome and injecting it into our embryonic hardware
would be very easy to do,” he says. Many would even consider this type of
electronics as a new species. What else do you call something that has cells and
genes, can evolve, replicate and heal itself?

This new species would certainly make Mission Control’s job far easier. If
embryonics works as promised, spacecraft will be able to survive a loss of
communications with ground control, and maybe even outlive their designers on
Earth. “It would enable us to plan encounters successfully without ground
intervention in the loop,” Erickson says. “We would not worry as much 19 hours
before a close encounter.” Which is worth a lot to Erickson. For one, he says,
he would sleep better.