DARPA’s Polymorphous Computing Architectures program [1] is developing a revolutionary approach to implementing embedded computing systems that support reactive, multi-mission, multi-sensor, and in-flight retargetable missions, and that reduce the time needed for payload adaptation, optimization, and validation from years to days to minutes. The PCA program breaks the current development approach of “hardware first and software last” point solutions by moving beyond conventional computer hardware and software to flexible, “polymorphous” computing systems. A polymorphous computing system (chips, processing architecture, memory, networks, and software) will “morph” (take on or pass through varying forms or implementations) to best fit changing mission requirements, sensor configurations, and operational constraints during a mission, for changing operational scenarios, or over the lifetime of a deployed platform.
In current practice, a processor is typically selected to perform a particular class of processing, such as real-time data signal processing or, at the other extreme of the processing spectrum, cognitive reasoning. A corresponding spectrum of domain-specific processors, such as specialized data-intensive signal processors, general digital signal processors (DSP), general-purpose microprocessors, and the server-class devices, is then required to perform the complete range of mission processing. Through their ability to reconfigure resources and architectural elements to implement a broad range of architectural implementations, PCA systems can span this processing continuum with a single class of device. This is accomplished by dynamically reorganizing the PCA device’s processing elements or micro-architectural components to provide an optimized architectural implementation for each specific set of mission or system requirements.
To achieve this capability, the PCA program is implementing a family of novel malleable micro-architecture processing elements to include compute cores, caches, memory structures, data paths, network interfaces, network fabrics with incremental instructions, OS, and network protocols. To support the use of these polymorphous computing systems, the program is creating a model-based software framework for reactive monitoring, optimization, modeling, resource negotiation and allocation, regeneration, and verification. A set of measurement metrics are being developed to support processing system design and optimization; these include size, weight, energy, performance, and time (SWEPT). Finally, the PCA program is establishing benchmark and standards groups that are creating community standards to enable broad application and commercial support of PCA program developments. The Morphware Stable Interface described in this document is a product of this latter activity.