Hi AusEire,
I did draft a response but it seems to have fallen down the back of the filing cabinet.
"
We fully utilised the NPU and AI models for the 16-bit RAW files. A snap will collect twenty shots of the background and users can use up to four cameras for the Ultra model simultaneously," the VP said.
DSLR cameras have large sensors that allow them to take RAW images without noise, but up to now, for smartphones cameras, this has been difficult to emulate in low-light settings as they pack smaller sensors. But with the computational power offered from the NPU on the Galaxy S22 series, their 16-bit RAW files match those taken by DSLR cameras, according to Cho.
"Professionals today want RAW files from their smartphones.
Akida can utilize up tp 4-bit weights (derived from the library models) and actuations (derived from the camera pixels). It would need a major rejigging of the NPU architecture to include a 16-bit MAC capability, and would significantly diminish speed and power advantages.
#################################################################
We looked at Samsung on 12 March (@Rocket77)
#5,093
This related to their 3D tech where they are building the neuron equivalent of the human brain. Their 3D tech is interesting as they stack MemRistors vertically on top of each other.
Hi Rocket577,
The video refers to memRistors a few times.
I looked at Samsung in October last year in the other place:
"Samsung appear to be using memristors in their 3-D NN:
US2021319293A1 NEUROMORPHIC DEVICE AND OPERATING METHOD OF THE SAME
https://worldwide.espacenet.com/pat...456/publication/US2021319293A1?q=US2021319293
an operating method of a neuromorphic device including at least one three-dimensional synaptic array having a plurality of layers, the operating method comprises upon receiving a plurality of input signals independent of each other, performing an artificial neural network computation on the input signals in a plurality of resistive memristor elements of a layer corresponding to a word line selection signal, as a result of the artificial neural network computation, outputting a plurality of output currents to neuron circuits through bit lines, summing the output currents to output a summed output current as an output voltage when it is more than a predetermined threshold and wherein the three-dimensional synaptic array performs an artificial neural network computation independently for each layer activated by the word line selection signal.
1 . A neuromorphic device, comprising:
at least one synaptic array,
wherein the synaptic array includes:
a plurality of input lines extending in a first direction and receiving input signals independently of each other from a plurality of axon circuits connected respectively thereto;
a plurality of bit lines extending in a second direction orthogonal to the first direction and outputting output signals independently of each other;
a plurality of cell strings, each including at least two resistive memristor elements and a string select transistor connected in series in a third direction between any one of the plurality of input lines and any one of the plurality of bit lines;
a plurality of electrode pads stacked while being spaced apart from each other in the third direction between the plurality of input lines and the plurality of bit lines, and connected to the string select transistor and the at least two resistive memristor elements;
a decoder configured to apply a string selection signal or a word line selection signal to each of the plurality of electrode pads; and
a plurality of neuron circuits, each being connected to one of the bit lines connected to one of the cell strings, summing the output signals, converting, and outputting a summed signal when it is more than a predetermined threshold,
wherein the synaptic array performs an artificial neural network computation on the input signals in the resistive memristor elements of a layer activated by the word line selection signal in at least one cell string to which the string selection signal is applied.
View attachment 3390
View attachment 3391
View attachment 3392
View attachment 3393
And they've got a natty arrangement for the synaptic array string (see Fig 6)
Opinions and research expressed are my own as an unlicensed individual. External links are not endorsed. Do your own research or consult a licensed financial advisor.