Limits...
Implementation of olfactory bulb glomerular-layer computations in a digital neurosynaptic core.

Imam N, Cleland TA, Manohar R, Merolla PA, Arthur JV, Akopyan F, Modha DS - Front Neurosci (2012)

Bottom Line: Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits.The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons.Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

View Article: PubMed Central - PubMed

Affiliation: Computer Systems Lab, Department of Electrical and Computer Engineering, Cornell University Ithaca, NY, USA.

ABSTRACT
We present a biomimetic system that captures essential functional properties of the glomerular layer of the mammalian olfactory bulb, specifically including its capacity to decorrelate similar odor representations without foreknowledge of the statistical distributions of analyte features. Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits. The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons. This configuration generates functional transformations comparable to those observed in the glomerular layer of the mammalian olfactory bulb. Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

No MeSH data available.


Related in: MedlinePlus

Normalization of the total activity of mitral cells via the ET/sSA network. (A) Illustration of the lateral connectivity of sSA cells. Glomeruli receive inputs from the axons of multiple sSA neurons located at a characteristic distribution of distances away. These axons branch extensively, although for clarity this is not shown. The sSA neurons receive synaptic input from ET cells in a small number of immediately neighboring glomeruli (depicted as the gray dendrites of the black sSA cells). Figure adapted from Cleland (2010). (B) Crossbar connectivity in the chip to replicate the normalization effects of the ET/sSA lateral network. Each sSA neuron receives input from ET cells from the same row. The PGe and ET neurons of each of the 48 columns receive input from 10 sSA neurons, most of which are located in neighboring rows, but some of which are more than five rows away. Spikes generated by an sSA neuron are routed to its axon in the crossbar which synapses with the PGe and ET cells of 10 different glomeruli. The outputs of the PGe and ET neurons also are routed to their axons in order to make the appropriate inter- and intra-glomerular connections (see Figure 1).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3368244&req=5

Figure 6: Normalization of the total activity of mitral cells via the ET/sSA network. (A) Illustration of the lateral connectivity of sSA cells. Glomeruli receive inputs from the axons of multiple sSA neurons located at a characteristic distribution of distances away. These axons branch extensively, although for clarity this is not shown. The sSA neurons receive synaptic input from ET cells in a small number of immediately neighboring glomeruli (depicted as the gray dendrites of the black sSA cells). Figure adapted from Cleland (2010). (B) Crossbar connectivity in the chip to replicate the normalization effects of the ET/sSA lateral network. Each sSA neuron receives input from ET cells from the same row. The PGe and ET neurons of each of the 48 columns receive input from 10 sSA neurons, most of which are located in neighboring rows, but some of which are more than five rows away. Spikes generated by an sSA neuron are routed to its axon in the crossbar which synapses with the PGe and ET cells of 10 different glomeruli. The outputs of the PGe and ET neurons also are routed to their axons in order to make the appropriate inter- and intra-glomerular connections (see Figure 1).

Mentions: The dense lateral network formed by ET and sSA cells has small-world-like connectivity (Cleland et al., 2007; Cleland, 2010; Figure 6A). This enables the integration of activity across the glomerular layer at a fraction of the energy and volume costs that would have been incurred with all-to-all connections. We followed this insight to configure an ET/sSA network in our chip (Figure 6B). The projection topology was sparse – a total of 10 sSA cells projected axons to a given glomerular column – and reflected a scaled version of the actual connectivity profile observed in the mouse olfactory bulb (Aungst et al., 2003). While the physical proximity of glomerular circuits on the chip is computationally irrelevant, for illustrative purposes we utilized their locations to organize the distribution of connectivity to resemble that of the ET/sSA network. Specifically, around 50% of the 10 sSA inputs arose from glomerular columns located more distant than two rows away in the two-dimensional neuron array, while around 20% arose from columns more than five rows away. An sSA axon branched to contact 10 different columns while its dendrite received input from the ET cells from the same row. We determined appropriate synaptic weights for the ET/sSA network by sweeping through a range of parameters and measuring the normalized Euclidean distance between the chip output and a z score normalization term produced by the same input pattern (Cleland et al., 2007). The z score was computed as [zi = (xi − μx)/σx], where μx was the mean of the synaptic inputs and σx was the standard deviation. The synaptic weights that minimized the normalized Euclidean distance were used.


Implementation of olfactory bulb glomerular-layer computations in a digital neurosynaptic core.

Imam N, Cleland TA, Manohar R, Merolla PA, Arthur JV, Akopyan F, Modha DS - Front Neurosci (2012)

Normalization of the total activity of mitral cells via the ET/sSA network. (A) Illustration of the lateral connectivity of sSA cells. Glomeruli receive inputs from the axons of multiple sSA neurons located at a characteristic distribution of distances away. These axons branch extensively, although for clarity this is not shown. The sSA neurons receive synaptic input from ET cells in a small number of immediately neighboring glomeruli (depicted as the gray dendrites of the black sSA cells). Figure adapted from Cleland (2010). (B) Crossbar connectivity in the chip to replicate the normalization effects of the ET/sSA lateral network. Each sSA neuron receives input from ET cells from the same row. The PGe and ET neurons of each of the 48 columns receive input from 10 sSA neurons, most of which are located in neighboring rows, but some of which are more than five rows away. Spikes generated by an sSA neuron are routed to its axon in the crossbar which synapses with the PGe and ET cells of 10 different glomeruli. The outputs of the PGe and ET neurons also are routed to their axons in order to make the appropriate inter- and intra-glomerular connections (see Figure 1).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3368244&req=5

Figure 6: Normalization of the total activity of mitral cells via the ET/sSA network. (A) Illustration of the lateral connectivity of sSA cells. Glomeruli receive inputs from the axons of multiple sSA neurons located at a characteristic distribution of distances away. These axons branch extensively, although for clarity this is not shown. The sSA neurons receive synaptic input from ET cells in a small number of immediately neighboring glomeruli (depicted as the gray dendrites of the black sSA cells). Figure adapted from Cleland (2010). (B) Crossbar connectivity in the chip to replicate the normalization effects of the ET/sSA lateral network. Each sSA neuron receives input from ET cells from the same row. The PGe and ET neurons of each of the 48 columns receive input from 10 sSA neurons, most of which are located in neighboring rows, but some of which are more than five rows away. Spikes generated by an sSA neuron are routed to its axon in the crossbar which synapses with the PGe and ET cells of 10 different glomeruli. The outputs of the PGe and ET neurons also are routed to their axons in order to make the appropriate inter- and intra-glomerular connections (see Figure 1).
Mentions: The dense lateral network formed by ET and sSA cells has small-world-like connectivity (Cleland et al., 2007; Cleland, 2010; Figure 6A). This enables the integration of activity across the glomerular layer at a fraction of the energy and volume costs that would have been incurred with all-to-all connections. We followed this insight to configure an ET/sSA network in our chip (Figure 6B). The projection topology was sparse – a total of 10 sSA cells projected axons to a given glomerular column – and reflected a scaled version of the actual connectivity profile observed in the mouse olfactory bulb (Aungst et al., 2003). While the physical proximity of glomerular circuits on the chip is computationally irrelevant, for illustrative purposes we utilized their locations to organize the distribution of connectivity to resemble that of the ET/sSA network. Specifically, around 50% of the 10 sSA inputs arose from glomerular columns located more distant than two rows away in the two-dimensional neuron array, while around 20% arose from columns more than five rows away. An sSA axon branched to contact 10 different columns while its dendrite received input from the ET cells from the same row. We determined appropriate synaptic weights for the ET/sSA network by sweeping through a range of parameters and measuring the normalized Euclidean distance between the chip output and a z score normalization term produced by the same input pattern (Cleland et al., 2007). The z score was computed as [zi = (xi − μx)/σx], where μx was the mean of the synaptic inputs and σx was the standard deviation. The synaptic weights that minimized the normalized Euclidean distance were used.

Bottom Line: Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits.The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons.Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

View Article: PubMed Central - PubMed

Affiliation: Computer Systems Lab, Department of Electrical and Computer Engineering, Cornell University Ithaca, NY, USA.

ABSTRACT
We present a biomimetic system that captures essential functional properties of the glomerular layer of the mammalian olfactory bulb, specifically including its capacity to decorrelate similar odor representations without foreknowledge of the statistical distributions of analyte features. Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits. The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons. This configuration generates functional transformations comparable to those observed in the glomerular layer of the mammalian olfactory bulb. Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

No MeSH data available.


Related in: MedlinePlus