These researchers have found another use for an SNN Chip:
“1 Introduction
Cloud services cause an exponentially growing traffic in data centers. This requires optical transceivers to operate at lower power and lower cost. Because digital signal processing (DSP) has high power consumption, recent research envisions replacing the DSP in part with an analog frontend with lower power
consumption….
“Conclusions
In this work, we have implemented an SNN on the mixed-signal neuromorphic hardware system BrainScaleS-2 (BSS-2). While we observed a hardware penalty slightly below 1dB at a BER of 2×10−3, the BSS-2 SNN outperforms a simulated linear equalizer with the same number of taps, thanks to nonlinear processing.
The presented results confirm that neuromorphic hardware can provide the reliability required by signal processing in optical transceivers.
Promising directions
for future research include reducing the hardware penalty, reducing the architec- tural complexity and increasing the intrin- sic speed of the SNN, and optimizing the input and output interfaces of the SNN. Furthermore, processing by the SNN must be parallelized. Also, the effective power consumption of neuromorphic signal pro- cessing should be analyzed and compared to digital processing.
In a future hardware implementation, spikes could be generated directly from the electrical signal coming from the photo diode, thereby replacing the power-hungry analog-to-digital converter (ADC). In this work, we have considered hard decision demapping. Another interesting direction is the design of an SNN equalizer/demap- per with soft output.”
About Brainscale:
4 BrainScaleS-2 System
We deploy our SNN equalizer/demapper on the accelerated mixed-signal neuromor phic BSS-2 system, developed at Heidel- berg University [5] in Germany.
On its analog neural network core, it emulates up to 512 LIF neurons and 128 k synapses in analog circuits in parallel and in continu- ous time. Hardware synapses have 6-bit weights, which can be configured as in- hibitory (negative sign) or excitatory (pos- itive sign). The neurons communicate via digital spike events. Each neuron is parameterized individually.
My opinion only DYOR
FF
AKIDA BALLISTA