Verification and Design Methods for the BrainScaleS Neuromorphic Hardware System

  • 2020-03-25 15:48:54
  • Andreas GrĂ¼bl, Sebastian Billaudelle, Benjamin Cramer, Vitali Karasenko, Johannes Schemmel
  • 1


This paper presents verification and implementation methods that have beendeveloped for the design of the BrainScaleS-2 65nm ASICs. The 2nd generationBrainScaleS chips are mixed-signal devices with tight coupling betweenfull-custom analog neuromorphic circuits and two general purposemicroprocessors (PPU) with SIMD extension for on-chip learning and plasticity.Simulation methods for automated analysis and pre-tapeout calibration of thehighly parameterizable analog neuron and synapse circuits and forhardware-software co-development of the digital logic and software stack arepresented. Accelerated operation of neuromorphic circuits and highly-paralleldigital data buses between the full-custom neuromorphic part and the PPUrequire custom methodologies to close the digital signal timing at theinterfaces. Novel extensions to the standard digital physical implementationdesign flow are highlighted. We present early results from the first full-sizeBrainScaleS-2 ASIC containing 512 neurons and 130K synapses, demonstrating thesuccessful application of these methods. An application example illustrates thefull functionality of the BrainScaleS-2 hybrid plasticity architecture.


Quick Read (beta)

loading the full paper ...