Recent Projects
For the full list of projects please check the CV.
OISMA is unconventional AI Computing Architecture utilizing On-the-fly In-memory Stochastic Multiplication Arrays for Digital In-Memory Computing.
Triangular Input Movement Systolic Array for Convolutional Workloads. TrIM eliminates the memory access’ overhead, associated with Convolutional workloads, reducing total numbers of memory accesses by an order of magnitude.
Diagonal-Input Permutated-Weights Systolic Array for Matrix-Multiplications. DiP doubles the energy-efficiency per area in comparison to Google TPU-like architectures by eliminating the associated interfacing synchronization logic (FIFOs).
TXL investigates new AI computing concepts for edge classification by utilizing a runtime analogue template matching approach. Analogue Content-Addressable Memory (ACAM) systems are proposed to replace fully connected layers of CNNs.
Memristors Characterization Platform. A collaboration with Imperial College London (UK) on building a characterization chip of a million RRAM devices.