72
3 SAMPA Chip Implementation
sync packets are the same size and format as a packet header, but the information in
it is fixed. As this keeps the links toggling while the device is idle, it will consume
some extra power. An option would have been to power off the serial links whenever
there are no more data to send and when a new packet is ready to be transmitted, a
sync packet would be sent first to enable the reviving side to synchronize to the data
stream. Currently power-off is only done for a serial link when it is not enabled.
DC balancing of the output serial links is not implemented in the device, as it is
not a requirement for interfacing with the GBTx. In other applications that do not use
the GBTx, the DC balancing might be of benefit to provide better signal integrity.
As the native word format of the SAMPA is 10-bit based, the best option would be
to use a 5b/6b encoding which would increase the overall bandwidth usage by 20 %.
The existing sync packets would not be necessary any more as the DC balancing
code has unused codes that can be used for link synchronization.
To reduce cost and development time, it is common to produce only one or maybe
a couple of different versions of a front-end card to mate with the detector pads.
However, the detector might have several different mappings of physical pads to
input channel on the front-end card, depending on the position in the detector. As the
serial links are independent of each other, the data from one chip can potentially be
sent to two or more upstream devices for further processing of the data for instance for
event reconstruction. If it is important to the detector to have specific input pads arrive
at specific upstream devices, independent of the mapping, the SAMPA provides a
way of remapping which input channels are presented to each serializer.
Due to the setup with a fixed set of channels available to each serializer, the
SAMPA currently provides no way of load balancing between the serializer in case
a set of channels for one serializer has a higher occupancy than the set for another.
This could cause unnecessary overflow of buffers, even though there is available
bandwidth on other serial links. If all serial links would have access to read from
all of the channels, it will provide a more equal load across the serial links. Each
link would then iterate through its subset and whenever all the channels in its subset
is empty, it picks another one not in its subset and which is non-empty and not
already being read. Normally it would send sync packets in this case. This method
will however cause the predictability of which channel will arrive next on which link
to be lost. It will also render the option to re-order which channels connect to which
serial link mostly pointless.
3.2.6.2 Daisy Chained Readout
For detectors with very low data rates, specifically the MCH, a daisy chaining option
is implemented. This lets multiple devices share a single serial link, increasing the
number of devices that can be read out by a single GBTx. Each device is provided with
an extra data input port and two data-control signals for this purpose. A connection
diagram can be seen in Fig. 3.20.
The upstream device is set up to run with one serial downlink and the busy out
signal from the downstream device is connected to the busy in of the upstream device.
3 SAMPA Chip Implementation
sync packets are the same size and format as a packet header, but the information in
it is fixed. As this keeps the links toggling while the device is idle, it will consume
some extra power. An option would have been to power off the serial links whenever
there are no more data to send and when a new packet is ready to be transmitted, a
sync packet would be sent first to enable the reviving side to synchronize to the data
stream. Currently power-off is only done for a serial link when it is not enabled.
DC balancing of the output serial links is not implemented in the device, as it is
not a requirement for interfacing with the GBTx. In other applications that do not use
the GBTx, the DC balancing might be of benefit to provide better signal integrity.
As the native word format of the SAMPA is 10-bit based, the best option would be
to use a 5b/6b encoding which would increase the overall bandwidth usage by 20 %.
The existing sync packets would not be necessary any more as the DC balancing
code has unused codes that can be used for link synchronization.
To reduce cost and development time, it is common to produce only one or maybe
a couple of different versions of a front-end card to mate with the detector pads.
However, the detector might have several different mappings of physical pads to
input channel on the front-end card, depending on the position in the detector. As the
serial links are independent of each other, the data from one chip can potentially be
sent to two or more upstream devices for further processing of the data for instance for
event reconstruction. If it is important to the detector to have specific input pads arrive
at specific upstream devices, independent of the mapping, the SAMPA provides a
way of remapping which input channels are presented to each serializer.
Due to the setup with a fixed set of channels available to each serializer, the
SAMPA currently provides no way of load balancing between the serializer in case
a set of channels for one serializer has a higher occupancy than the set for another.
This could cause unnecessary overflow of buffers, even though there is available
bandwidth on other serial links. If all serial links would have access to read from
all of the channels, it will provide a more equal load across the serial links. Each
link would then iterate through its subset and whenever all the channels in its subset
is empty, it picks another one not in its subset and which is non-empty and not
already being read. Normally it would send sync packets in this case. This method
will however cause the predictability of which channel will arrive next on which link
to be lost. It will also render the option to re-order which channels connect to which
serial link mostly pointless.
3.2.6.2 Daisy Chained Readout
For detectors with very low data rates, specifically the MCH, a daisy chaining option
is implemented. This lets multiple devices share a single serial link, increasing the
number of devices that can be read out by a single GBTx. Each device is provided with
an extra data input port and two data-control signals for this purpose. A connection
diagram can be seen in Fig. 3.20.
The upstream device is set up to run with one serial downlink and the busy out
signal from the downstream device is connected to the busy in of the upstream device.