The 1124 Kasli 2.0 Carrier card is an 8hp EEM module, designed to run ARTIQ kernels sent from a host machine over the network. It supports up to 12 EEM connections to other EEM cards in the ARTIQ-Sinara family and up four SFP connections, used for comunications with other carriers and/or Ethernet.
4 SFP 6Gb/s slots are provided. One may be used for Ethernet, which supports communication with a host machine. Remaining slots can be used by the ARTIQ Distributed Real-Time Input/Output (DRTIO) system, which allows for the use of additional core devices (e.g. Kasli 2.0, Kasli-SoC) as satellite cards, capable of running subkernels or distributing commands from the \mbox{DRTIO} master.
DS181\footnote{\label{ds181}\url{https://docs.xilinx.com/v/u/en-US/ds181\_Artix\_7\_Data\_Sheet}}. These figures account for the insertion loss of the RF transformer (TC2-1TX+\footnote{\label{rf_trans}\url{https://www.minicircuits.com/pdfs/TC2-1TX+.pdf}}).
Power is to be supplied through the barrel connector in the front panel, size 5.5 mm OD, 2.5 mm ID, and is passed on to daughtercards through the EEM connections. Locking barrel connectors are supported.
Kasli 2.0 features an XC7A100T-3FGG484E Xilinx Artix-7 FPGA to facilitate reconfigurable high-speed real-time control of inputs and outputs. Most commonly, this FPGA is flashed with binaries compiled from the ARTIQ (Advanced Real-Time Infrastructure for Quantum physics) control system, which equips the carrier board with specialized gateware for handling other Sinara EEMs and an on-FPGA CPU for running ARTIQ experiment \mbox{kernels}.
ARTIQ is open-source and can be found in the repository \url{https://github.com/m-labs/artiq}. Orders of Sinara hardware are normally accompanied with precompiled binaries. Long-term support for ARTIQ systems can also be purchased.
DRTIO is the time and data transfer system that allows ARTIQ RTIO channels to be distributed among several core device carrier boards, synchronized and controlled by a central core device. The system itself is more fully described in the ARTIQ documentation\footnote{\label{manual-drtio}\url{https://m-labs.hk/artiq/manual/drtio.html}}. With ARTIQ firmware/gateware, supported core devices, including Kasli 2.0, can take one of three roles:
The DRTIO master is unique in a DRTIO system. It requires a direct network connection to the host machine. It may make downstream connections to satellites. It controls its own local RTIO channels and the downstream DRTIO satellite(s).
\item\textbf{Satellite}\\
Any other core devices in a DRTIO system are DRTIO satellites. They require an upstream connection to one other core device, master or satellite, through which communications will ultimately be chained to the master. They may make further downstream connections to other satellites. They may control RTIO channels through subkernels or simply pass on communications from the master.
When run in a non-distributed ARTIQ configuration, with a single central core device but no satellites, that core device is instead known as standalone.
Communication between devices is implemented using 1000Base-T small form-factor pluggable (SFP) interfaces. Four are available on the Kasli 2.0. Appropriate SFP transceivers must be plugged inside the corresponding SFP cages. Each SFP connector possesses an indicator LED.
An Ethernet-capable SFP transceiver should be inserted into the \texttt{SFP0} slot. Typically, a 10000Base-X RJ45 SFP module is used, with an network-connected Ethernet cable attached to the module.
\item\textbf{Satellite}\\
The \texttt{SFP0} port should be connected to one of the free SFP slots on an upstream core device, using a cable connection with SFP transceivers.
Kasli 2.0 supports up to 3 DRTIO satellite connections per device. Any of the 3 downstream SFP ports (i.e. \texttt{SFP1}, \texttt{SFP2}, \texttt{SFP3}) may be used.
The RTIO clock is typically synthesized by the Si5324 clock multiplier and distributed by the ADCLK948 clock fanout buffer to both the FPGA and the MMCX connectors. Alternatively, an external reference can be supplied through the front panel SMA connector. It is then buffered in the FPGA and sent to the Si5324 for clock synthesis. Kasli 2.0 supports a set of RTIO clock options:
The clock synthesizer may also be bypassed, using the \texttt{ext0\char`_bypass} option, which will accept a RTIO clock directly supplied to the SMA connector. The resulting signal is then routed to both the RTIO system and downstream satellites.
Clocking options in a running system should be configured by setting the value of the \texttt{rtio\char`_clock} key to the desired configuration through \texttt{artiq\char`_coremgmt}. For example, a RTIO frequency of 125MHz will be synthesized from an external 10 MHz signal after issuing the following command:
The RTIO clock is recovered from the SFP transceiver connected to the upstream device. The resulting signal is then cleaned up by the Si5324 and routed to both the RTIO system and downstream satellites.
Kasli 2.0 can be configured to use WRPLL, a clock recovery method making use of White Rabbit's DDMTD (Digital Dual Mixer Time Difference) and the card's Si549 oscillators, both to lock the main RTIO clock and to lock satellite clocks to master.
Kasli 2.0 supplies three user LEDs for debugging purposes. Two are located on the front panel. The third is located on the PCB itself, beside the SFP cage. An additional ERR LED on the front panel is used by ARTIQ firmware to indicate a runtime panic.
Instead of directly emitting RTIO events, sequences of RTIO events can be recorded in advance and stored in the local SDRAM. The event sequence can then be replayed at a specified timestamp. This is of special advantage in cases where RTIO events are too closely placed to be generated as they are executed, as events can be replayed at a higher speed than the on-FPGA CPU alone is capable of.
The following example records an LED event sequence and replays it twice consecutively using \texttt{CoreDMA}. When run, the LED should blink twice.
\subsection{Dataset manipulation with core device cache}
Experiments may require values computed or found in previously executed kernels. To avoid invoking an RPC or sacrificing the pre-computation in \texttt{prepare()} stage, data can be stored in the core device cache.
The following code snippets describe two experiments, in which the data from the first experiment is cached. The data is then retrieved and printed in hexadecimal form in the second experiment.
