Programming through biology

Getting it together Strand displacement computing

Inputs in strand-displacement computing are free-floating single nucleic acid strands. Logic gates are complexes of two or more strands, one of which is the potential output signal. Sticky overhanging tabs (or ‘toeholds’) on the gates allow passing input signals to latch-on (see upper diagram below). If an input signal has a base-pair sequence complementary to a gate, it binds to it, displacing the output strand.. The free- floating output can then trigger a downstream gate so the circuit operates like a cascade (lower diagram below). Programming involves choosing the  specific base sequences that make-up the different gates, and the number of types of gate and input strands.

Synthetic Biology makes the switch Building organismic electronics

One aspect of biological computing has recently shown that intestinal bacteria and even DNA can form the basis of ‘logic gates’ - devices that can be switched between two opposing states - using chemical triggers.

Biological logic gates could, think researchers at the Department of Bioengineering and the Faculty of Natural Sciences at Imperial College London, be used as ‘processors’ in microscopic ‘biological computers’. They would take form of sensors designed to detect and address pollutants and toxic substances in at-risk environmental situations, such as places with a risk of toxin leakage.

The four-year-long project - ‘Engineering modular and orthogonal genetic logic gates for robust digital-like synthetic biology’ - built a type of biological logic gate called an AND gate from an E Coli sample with modified DNA, which reprogrammed it to perform a ‘near-digital’ on-and-off switching process when stimulated by chemicals.

“In the test system we used small, simple, naturally occurring sugars as primary signals to turn on the genes,” says Professor Martin Buck of Imperial College. “These sugars can be used by bacteria for growth, and we employed the regulatory systems that control the sugar utilisation genes as primary building blocks.”

Next, a NOT gate was created, which was then combined with the AND gate to produce the more complex NAND gate. The research will attempt to develop more complex circuitry that comprises multiple logic gates. One challenge here is linking multiple logic gates together in a way similar to how electronic logic gates connect together to enable complex processing.

“Biological computing devices may include sensors that swim inside arteries to detect the build-up of plaque, and even to deliver medications to affected zones,” says Professor Buck. Other applications posited include sensors capable of detecting and destroying cancer cells.

The interdisciplinary team - which also included Professor Richard Kitney of Imperial’s Centre for Synthetic Biology and Innovation and Department of Bioengineering, Dr Nicholas Joly of the Faculty of Natural Sciences, and Baojun Wang of the Faculty of Natural Science - say that its demonstrations show that biological logic gates can be linked: this suggests the possibility of modular biological processors.

“The difference with the Imperial project is that our devices are predictable and robust, and have been shown to work in differing environments,” says Professor Kitney. “Discovering that we were able to control a naturally-occurring biological mechanism in the way we wanted to was something of a surprise. Establishing a biologically-based control system means that you can override cellular functions and start to look at ways to constraining a cell so that it operates in a desired way.”

“The next research stage could lead to a type of circuitry for processing information,” suggests Professor Buck, “complex biological circuitry processing information using chemicals, in the same way that our body uses them to process and store information. Bacterial and biological-based computers have to be seen as fundamentally different systems to silicon-based devices, and may never match the latter’s instruction execution rate. They may match in memory, but these will much slower than electronic computers in speed of operation… [However] they may well excel in dealing with problems where it is unclear what instructions to give; complex problems could instead be evolved to an answer.”

Bio-gates have more advantages over other synthesised techniques, says Professor Kitney: “They are better suited to functioning in harsh or unpredictable environments, and are able to draw energy from their local environments.”

By James Hayes, Technology Features Editor.

Chips off of the old block packet-switched brain models, and cognitive computing

University of Manchester and IBM US have, in parallel projects, developed cognitive computing chips to create working models of human brain-functions. The idea of both projects is to imitate brain abilities, such as sensation, perception, action, interaction, and cognition using high-performance computers, then see how insights can inform the development of next-generation computer systems.

Computer scientists at the University of Manchester say they have created chips based on ARM processor technology which models the human brain. Professor Steve Furber and his team from the University’s School of Computer Science have custom built the architecture for a supercomputer platform called Spinnaker (Spiking Neural Network Architecture); its purpose is to serve as an integral tool for neuroscientists, psychologists, and doctors as they understand complex brain injuries, diseases, and other conditions. For research neuroscientists, Spinnaker could help build large-scale models to verify hypotheses of neural functions; scientists can then test theories on scales and at speeds which Furber claims to be unachievable.

“I’ve been in microelectronics computing for 30 years. I’ve seen computers get faster, but they still struggle to do simple things, but to see high-performance computers act like human brains it a big step forward,” says Furber. “We don’t know how the brain works as an information-processing system, but we need to find out. Spinnaker will help gain key insights into brain activity, and patients will benefit from this.”

University of Manchester was selected to design the system architecture and received half of the £5m Engineering and Physical Sciences Research Council (EPSRC) grant; the rest is used at the universities of Southampton, Cambridge, and Sheffield, to manage the architecture, software and application sides. The team members at Manchester range from faculty academics to electrical, computer science, and psychology PhD students.

The starting point for the Spinnaker project is the fact that the basic way high-performance computers (HPC) work already has similarities with brain functions.

Modelling the 100 billion neurons with 1,000 million connections typical to the human brain, Spinnaker allows the neurons to release spikes which are relayed as minute electrical signals. Each impulse is modelled in the system as a ‘packet’ of IP data, comparable to the way the Internet transports information using IP as the principal communications protocol that used for relaying datagrams (packets) across connected networks using the Internet Protocol Suite. The ‘packet’ is then transmitted to all connected neurons which are represented by basic equations. These equations are then processed in real-time by software on the ARM chips. Each chip contains 18 ARM processors integrated in a single 19mm square package, with a second microchip that provides substantial memory using 3D system-in-package (SiP) technology.

“One of the measures that HPC folk use is the ‘bisection bandwidth’,” explains Professor Furber. “This, simply, is a measure of the bandwidth between two logical halves on the machine - if the machine were divided into two equal halves, how well would the two halves be connected? In Spinnaker, the bisection bandwidth amounts to five billion packets per second in each direction. It isn’t a big number in bits/second (250Gbps each way) as the packets are very small, but for brain-modelling small packets are all that is needed, so this number of packets is very high.”

Spinnaker systems have four nodes with 72 ARM processors, which equals to 18 processors per node. Furber says that the next-generation boards will have 48 nodes with 864 ARM processors; these boards will then be assembled into systems with up to a million ARM processors. The aspect of Spinnaker architecture that is most fixed and determined is the way the models on various ARM processors communicate with each other, by sending small packets of information. In a neural model each packet can represent the action potential or spike from a single neuron, and we can route that packet to the many thousands of other neurons in biological real time.

The University of Manchester team claims the electronic connections in Spinnaker convey these spikes much quicker and in fewer connections (five billion packets a second) compared with the biological connections in the brain. The new circuit board with 48 Spinnaker chips will be available in 2012.

Meanwhile, IBM has announced its own take on experimental cognitive computer chips to recreate the connections between spiking neurons and synapses in the brain through advanced technologies. IBM US researchers Paul Merolla’s and John Arthur’s project has some resonance with the work underway at University of Manchester, but the applications IBM foresees for its The Synapse (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) project was a result of a grand challenge initiative at IBM Research - Almaden in 2006. Every year IBM management asks for the next big idea, the winner receives resources and a two year trial period. In 2008 IBM and the universities of California, Merced, Columbia, Cornell and Wisconsin, Madison were awarded $21m from DARPA (Defence Advanced Research Projects Agency). IBM collaborated with the universities as they are well known in neuroscience, supercomputing, and nanotechnology; the team varied in expertise, ranging from psychiatrists, neurologists, chip designers, mathematicians to simulation experts.

The cognitive computers are expected to learn through experiences, find correlations, create hypotheses and remember and learn from outcomes, therefore mimicking the brain structural and synaptic plasticity. IBM has two working prototype designs; both cores were fabricated in 45nm SOI-CMOS, and contain 256 neurons. One core contains 262,144 programmable synapses, and the other contains 65,536 learning synapses. The goal of Synapse is to create a system that not only analyses complex information from multiple sensory modalities, but also rewires itself to interact with the environment.

Cognitive computing systems have the potential to apply brain-like functionality to a range of high-end applications. They can, for example, monitor the environment by keeping records of global water supply, and report metrics such as temperature, pressure, wave height, and ocean tide. Cognitive computing also extends to the healthcare sector, for spotting abnormal scenarios on roads and alerting law enforcements, and recognising opportunities, risk or anomalies in the financial sector.

“These chips are step in the evolution of computers from calculators to learning systems,” says project leader for IBM Research Dhamendra Modha.

By Aasha Bodhani, Assistant Editor.