Bioengineers at the University of Stanford have developed a new type of circuit board modeled on the human brain – perhaps bridging new gaps in computer science and mechatronics. Although the new development allows for a computer to operate thousands of times faster than a standard PC, it is still dwarfed by the power and capacity of the human brain. Nonetheless, this development may have opened a new chapter in bioengineering.
The new computer hardware developed at Stanford is capable of simulating, in real-time, a brain-worthy million neurons with billions of synaptic connections. The team at Stanford states that such an immense amount of computing power can be achieved by using only what is required to run an iPad.
Engineering computer models that have the ability to simulate the expansiveness and efficiency of the human is undoubtedly a complicated endeavor. For example, computer simulations of the cortex of a mouse are approximately 9,000 times slower than an organic system. In an article written for the Proceedings of the IEEE, Kwabena Boahen, associate professor of bioengineering at the University of Stanford, explained that a PC needs more than 40,000 times the power to complete the task. The Human Brain Project has stated its mission to simulate a human-scale cortex; however, it is predicted to need the energy of 250,000 households.
In the computer/brain study, scientists have produced a system they have dubbed, Neurogrid. Neurogrid is comprised of 16 “Neurocore” chips integrated together within a circuit board. Each Neurocore chip has a capacity of 65,536 neurons, which means that one circuit board can simulate over one million neurons. Considering the capacity of each neuron, billions of synaptic connections can be computer-simulated in real-time. Astonishingly, the application requires only 3 watts to run – the energy required to run a small light bulb. In comparison, a normal desktop PC handling twice the computing load uses several hundred watts and is 9,000 times slower than real-time.
This new development could solve the engineering problems for the desire to create advanced mechatronics for synthetic brain development in robotics. Moreover, it could aid scientist’s goal of better understanding the workings of the human brain and further bioengineering developments.
However, there is a rather unfortunate drawback – the cost. Each Neurogrid has a price tag of $40,000. Although this isn’t budget friendly, the team stated that they are currently using technology from 15 years ago. Furthermore, they explained that if they transfer to a modern fabrication means, they could cut the price 100-fold, to around $400. Even better though, they could increase the memory capacity of the Neurocore by two order of magnitude.
Notwithstanding of the system’s weak capacity compared to the human brain, operating a network of 80,000 times as many neurons only requires just three times the power of running the Neurogrid. Scientifically speaking, the future is bright for this developing application.
Scientists hope that the Neurogrid’s incredible efficiency may be able to be integrated to the mechanics of prosthetic limbs to allow for greater drive comparable to organic human motion. As well, there are hopes of eventually creating a system to be placed in the brain of a paralyzed human in order to act as a computer-to-brain translator. Nevertheless, the developments in bioengineering at Stanford show that even though creating a synthetic version the human brain is long away, the Neurogrid is a step in the right direction towards the future of bioengineering and creating brain-like computing power.
By: Alex Lemieux