36 lines
2.8 KiB
TeX
36 lines
2.8 KiB
TeX
%! TEX root = ../thesis.tex
|
|
|
|
\chapter{Outlook}
|
|
|
|
While all of the measurements could be calibrated they can still be fine tuned.
|
|
The found accuracy was \(\approx\) \SI{.33}{\%}.
|
|
To gain a better accuracy would require further iterations of the calibration process.
|
|
In the future it should also be possible to make these even more accurate, either by using a different method than the internally used polynomial of second degree.
|
|
This calculation could be of third degree or not even a polynomial.
|
|
|
|
The voltage distribution over the complete wafer was measured.
|
|
Using this distribution and V\(_\text{drop}\)s behavior at different loads could then be combined with the first iteration regulation model (SWRM).
|
|
This produced both the worst-case V\(_\text{drop}\) range as well as a set of formulas which could be used further.
|
|
These formulas allowed for regulating the \SI{1.8}{\volt} terminals.
|
|
Additionally a threshold current I\(_\text{thresh}\) could be observed.
|
|
This current limits the regulation mechanism, and results in still usable, but unregulated V\(_\text{drop}\).
|
|
This threshold current was also predictable with the theoretical considerations.
|
|
To further develop this mechanism, another, more complex, model (DWRM) was proposed.
|
|
The DWRM would allow for a more accurate regulation, which is specific to a given experiment and its chosen reticles.
|
|
This model would narrow down the worst-case scenario depicted by figures~\ref{fig:reg} and~\ref{fig:postreg}.
|
|
For that model to work, each experiment run on a HICANN wafer, would require a simulation of the distribution of voltage between the used reticles.
|
|
This would minimize the maximum difference in voltage drop considerably.
|
|
|
|
Furthermore the observed current threshold of \SI{81.3}{\ampere} is restricted by a internally used resistor value.
|
|
If the minimum resistance of the \SI{1.8}{\volt} generating circuit were to be decreased, the threshold would increase.
|
|
|
|
There are also external effects wich were not covered in this thesis.
|
|
One of those is that all systems are subject to temperature changes and therefore might not be stable or noisy.
|
|
Investigating this dependency would also allow for possible regulation mechanisms to compensate for changes in temperature.
|
|
|
|
Running a calibrated and regulating PowerIt inside a HICANN wafer system, would be the next step in testing the regulation mechanism.
|
|
The now regulated voltages could be resulting in more stable experiments.
|
|
It would also be feasable to now test the influence of different voltages on the wafers neuromorphic chips and their calculations.
|
|
Additionally to run an experiment on a HICANN wafer, a calibration is needed, and could also be influenced by the now regulated \SI{1.8}{\volt} voltages.
|
|
|
|
|