Integrated circuit CPU performance has come to the practical limits of clock-speed performance in large silicon-based IC design. Issues with power use and waste heat dissipation have limited the clock frequency of modern CPUs to 4GHz. Many groups are investigating carbon-based circuits. Though they promise perhaps a 200-fold improvement in clock speed over silicon-based circuits, this clock-speed improvement means Moore's Law will only be able to be upheld by transistor speed improvements for another 16 years-until 2028.

One easy way beyond the clock-cycle limitations lies in greater parallelism. The recent trend towards multiple cores in CPU design shows this movement has already begun in earnest. At first there seems to be no limit to the gains which can be made in this way but Amdahl's Law shows there are fundamental limitations which cause this approach to yield diminishing returns. We feel this is a problem not of implementation but rather is more fundamental, requiring a new computational paradigm.

Queral Networks (QNs) are a novel computational architecture designed, from the beginning, to circumvent the basic limitations of Amdahl's Law. We began with the premise that the von Neumann architecture is fundamentally limited with regard to parallelism due to its synchronous nature. To avoid Amdahl's Law, we believe one must make a fresh start designing a computational paradigm, making every design choice with parallelism as the paramount priority.

With QNs, we have taken an integrated approach to designing highly parallel systems. We draw inspiration from many fields: neuroscience, cognitive science, systems theory, complexity, genetic algorithms, artificial neural networks, computation, machine learning, and human factors.

QNs begin with the basic computational node, the Queron (a generalization of the neuron), allow it to connect to other querons following the rules of Kahn Process Networks, and add to that key restrictions to enforce modularity and reuse of these nodes. Following the best practices of software engineering and systems theory, QNs are designed to leverage the parallelism of neural networks while shining some light into their "black box". QNs allow complex solutions to be developed manually or trained by machine learning algorithms components of each can be reusued equally well.