400
P. A. Dananjaya et al.
obtained through the ex situ training process of a CNN. The predetermined kernel
weights are labelled as, W, as shown in Fig. 7. If the 3 × 3 kernel array to be
converted into the conductance values of the synapses in a single column in RRAM
crossbar array, consistent mapping rules are required. If the kernel weight can either
take positive or negative values, separating these values into two different columns
might be easier for the network to process, i.e., positive weights (marked in blue) and
negative weights (marked in pink). Similarly, for the input, X, it is divided into two
matrices one is positive (marked in green) and another is negated input, −X (marked
in orange). Moreover, each of the kernel weight must be connected with the correct
input node.
Once the weight matrix and input matrix values and arrangement are ready, the
weight values can be programmed into the RRAM synapse conductance, i.e., positive
weights occupy the top of the crossbar column while negative weights occupy the
bottom part. The corresponding conductance values (σ) can be mapped from the
Fig. 7 The weights and input activations used in the crossbar architecture of a neuromorphic core.
Different coloring corresponds to positive and negative weights and inputs. This figure is adapted
from [4]
P. A. Dananjaya et al.
obtained through the ex situ training process of a CNN. The predetermined kernel
weights are labelled as, W, as shown in Fig. 7. If the 3 × 3 kernel array to be
converted into the conductance values of the synapses in a single column in RRAM
crossbar array, consistent mapping rules are required. If the kernel weight can either
take positive or negative values, separating these values into two different columns
might be easier for the network to process, i.e., positive weights (marked in blue) and
negative weights (marked in pink). Similarly, for the input, X, it is divided into two
matrices one is positive (marked in green) and another is negated input, −X (marked
in orange). Moreover, each of the kernel weight must be connected with the correct
input node.
Once the weight matrix and input matrix values and arrangement are ready, the
weight values can be programmed into the RRAM synapse conductance, i.e., positive
weights occupy the top of the crossbar column while negative weights occupy the
bottom part. The corresponding conductance values (σ) can be mapped from the
Fig. 7 The weights and input activations used in the crossbar architecture of a neuromorphic core.
Different coloring corresponds to positive and negative weights and inputs. This figure is adapted
from [4]