Aspinity's RAMP (Reconfigurable Analog Modular Processor) technology is an analog processing platform that brings the versatility, performance, and consistency of digital neural processing systems into the lower-power analog domain to achieve the increasing levels of efficiency required for implementing sophisticated AI at the edge.
While it has long been recognized that analog can help us build much more power-efficient neural networks and AI systems –the implementation into deployable AI products has been stymied by the challenges typically associated with low-power analog circuit design such as susceptibility to manufacturing and environment variations, requirement for different fixed function analog circuits, and the lack of available high-precision analog memory that can be implemented at the compute elements.
RAMP enables a new level of power-efficiency for edge applications such as always-on sensing, generative AI, and myriad others with an analog neural processing technology that delivers:
- Near-zero power analog inferencing
- Scalability in network size and technology node
- High volume repeatable manufacturability
- Flexibility and programmability
Near-zero power neural acceleration:
The combination of ultra-low power analog circuits that are (1) co-located with, but separate from high-precision analog NVM, (2) immune to manufacturing and environmental variation, and (3) fully software programmable and configurable opens the door for RAMP technology to deliver a level of AI compute efficiency that is not addressed with today’s technology.
RAMP analog neural processing is extremely flexible and can:
- Be applied to a variety of NN topologies (CNNs, RNNs, etc)
- Scale up in NN size (number of parameters)
- Scale down in CMOS technology node (to 40nm or beyond)
As a result, RAMP technology is uniquely positioned to scale to 150+ TOPS/W to address the need for highly efficient NPUs (neural processing units) in mobile device chipsets that will enable the rapid growth of generative AI applications in mobile devices.
Near-zero power always-on edge sensing:
Because RAMP enables myriad functions in analog beyond inferencing, it is an extremely effective solution for ultra-low power always-on sensing at the edge. This functionality has been proven out in silicon with the development of Aspinity’s first product, the
AML100 analogML (analog machine learning) processor. The
AML100 is a fully analog processing platform that uses just 10's of µAs to process and classify raw, unstructured sensor data in the analog domain, keeping the MCU or other downstream digital processors asleep unless a relevant event is detected. The
AML100 reduces always-on AI system power by up to 100x, delivering a 20x or more improvement in battery life and dramatically reducing the amount of data transmitted to the cloud.
Learn more about the AML100