forked from M-Labs/artiq
1
0
Fork 0

doc: Add ZC706 to core device page

This commit is contained in:
architeuthidae 2024-08-20 15:55:54 +08:00 committed by Sébastien Bourdeauducq
parent 70d0f930c6
commit e38dc59656
2 changed files with 60 additions and 46 deletions

View File

@ -3,14 +3,14 @@ Core device
The core device is a FPGA-based hardware component that contains a softcore or hardcore CPU tightly coupled with the so-called RTIO core, which runs in gateware and provides precision timing. The CPU executes Python code that is statically compiled by the ARTIQ compiler and communicates with peripherals (TTL, DDS, etc.) through the RTIO core, as described in :doc:`rtio`. This architecture provides high timing resolution, low latency, low jitter, high-level programming capabilities, and good integration with the rest of the Python experiment code. The core device is a FPGA-based hardware component that contains a softcore or hardcore CPU tightly coupled with the so-called RTIO core, which runs in gateware and provides precision timing. The CPU executes Python code that is statically compiled by the ARTIQ compiler and communicates with peripherals (TTL, DDS, etc.) through the RTIO core, as described in :doc:`rtio`. This architecture provides high timing resolution, low latency, low jitter, high-level programming capabilities, and good integration with the rest of the Python experiment code.
While it is possible to use the other parts of ARTIQ (controllers, master, GUI, dataset management, etc.) without a core device, many experiments require it. While it is possible to use the other parts of ARTIQ (controllers, master, GUI, dataset management, etc.) without a core device, most use cases will require it.
.. _core-device-flash-storage: .. _configuration-storage:
Flash storage Configuration storage
------------- ---------------------
The core device contains some flash storage space which is used to store configuration data. It is one sector (typically 64 kB) large and organized as a list of key-value records, accessible either through :mod:`~artiq.frontend.artiq_mkfs` and :mod:`~artiq.frontend.artiq_flash` or, preferably in most cases, the ``config`` option of the :mod:`~artiq.frontend.artiq_coremgmt` core management tool (see below). Information can be stored to keys of any name, but the specific keys currently used and referenced by ARTIQ are summarized below: The core device reserves some storage space (either flash or directly on SD card, depending on target board) to store configuration data. The configuration data is organized as a list of key-value records, accessible either through :mod:`~artiq.frontend.artiq_mkfs` and :mod:`~artiq.frontend.artiq_flash` or, preferably in most cases, the ``config`` option of the :mod:`~artiq.frontend.artiq_coremgmt` core management tool (see below). Information can be stored to keys of any name, but the specific keys currently used and referenced by ARTIQ are summarized below:
``idle_kernel`` ``idle_kernel``
Stores (compiled ``.tar`` or ``.elf`` binary of) idle kernel. See :ref:`core-device-config`. Stores (compiled ``.tar`` or ``.elf`` binary of) idle kernel. See :ref:`core-device-config`.
@ -77,6 +77,35 @@ The same option is used to write ``boot.bin`` in ARTIQ-Zynq. Note that the ``boo
See also the full reference of :mod:`~artiq.frontend.artiq_coremgmt` in :ref:`Utilities <core-device-management-tool>`. See also the full reference of :mod:`~artiq.frontend.artiq_coremgmt` in :ref:`Utilities <core-device-management-tool>`.
.. _core-device-clocking:
Clocking
--------
The core device generates the RTIO clock using a PLL locked either to an internal crystal or to an external frequency reference. If choosing the latter, external reference must be provided (via front panel SMA input on Kasli boards). Valid configuration options include:
* ``int_100`` - internal crystal reference is used to synthesize a 100MHz RTIO clock,
* ``int_125`` - internal crystal reference is used to synthesize a 125MHz RTIO clock (default option),
* ``int_150`` - internal crystal reference is used to synthesize a 150MHz RTIO clock.
* ``ext0_synth0_10to125`` - external 10MHz reference clock used to synthesize a 125MHz RTIO clock,
* ``ext0_synth0_80to125`` - external 80MHz reference clock used to synthesize a 125MHz RTIO clock,
* ``ext0_synth0_100to125`` - external 100MHz reference clock used to synthesize a 125MHz RTIO clock,
* ``ext0_synth0_125to125`` - external 125MHz reference clock used to synthesize a 125MHz RTIO clock.
The selected option can be observed in the core device boot logs and accessed using ``artiq_coremgmt config`` with key ``rtio_clock``.
As of ARTIQ 8, it is now possible for Kasli and Kasli-SoC configurations to enable WRPLL -- a clock recovery method using `DDMTD <http://white-rabbit.web.cern.ch/documents/DDMTD_for_Sub-ns_Synchronization.pdf>`_ and Si549 oscillators -- both to lock the main RTIO clock and (in DRTIO configurations) to lock satellites to master. This is set by the ``enable_wrpll`` option in the :ref:`JSON description file <system-description>`. Because WRPLL requires slightly different gateware and firmware, it is necessary to re-flash devices to enable or disable it in extant systems. If you would like to obtain the firmware for a different WRPLL setting through AFWS, write to the helpdesk@ email.
If phase noise performance is the priority, it is recommended to use ``ext0_synth0_125to125`` over other ``ext0`` options, as this bypasses the (noisy) MMCM.
If not using WRPLL, PLL can also be bypassed entirely with the options
* ``ext0_bypass`` (input clock used directly)
* ``ext0_bypass_125`` (explicit alias)
* ``ext0_bypass_100`` (explicit alias)
Bypassing the PLL ensures the skews between input clock, downstream clock outputs, and RTIO clock are deterministic across reboots of the system. This is useful when phase determinism is required in situations where the reference clock fans out to other devices before reaching the master.
Board details Board details
------------- -------------
@ -88,20 +117,20 @@ All boards have a serial interface running at 115200bps 8-N-1 that can be used f
Kasli and Kasli-SoC Kasli and Kasli-SoC
^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^
`Kasli <https://github.com/sinara-hw/Kasli/wiki>`_ and `Kasli-SoC <https://github.com/sinara-hw/Kasli-SOC/wiki>`_ are versatile core devices designed for ARTIQ as part of the open-source `Sinara <https://github.com/sinara-hw/meta/wiki>`_ family of boards. All support interfacing to various EEM daughterboards (TTL, DDS, ADC, DAC...) through twelve onboard EEM ports. Kasli-SoC, which runs on a separate `Zynq port <https://git.m-labs.hk/M-Labs/artiq-zynq>`_ of the ARTIQ firmware, is architecturally separate, among other things being capable of performing much heavier software computations at high speeds on the board itself, but provides generally similar features to Kasli. Kasli itself exists in two versions, of which the improved Kasli v2.0 is now in more common use; the original Kasli v1.0 remains supported by ARTIQ. `Kasli <https://github.com/sinara-hw/Kasli/wiki>`_ and `Kasli-SoC <https://github.com/sinara-hw/Kasli-SOC/wiki>`_ are versatile core devices designed for ARTIQ as part of the open-source `Sinara <https://github.com/sinara-hw/meta/wiki>`_ family of boards. All support interfacing to various EEM daughterboards (TTL, DDS, ADC, DAC...) through twelve onboard EEM ports. Kasli is based on a Xilinx Artix-7 FPGA, and Kasli-SoC, which runs on a separate `Zynq port <https://git.m-labs.hk/M-Labs/artiq-zynq>`_ of the ARTIQ firmware, is based on a Zynq-7000 SoC, notably including an ARM CPU allowing for much heavier software computations at high speeds. They are architecturally very different but supply similar feature sets. Kasli itself exists in two versions, of which the improved Kasli v2.0 is now in more common use, but the original v1.0 remains supported by ARTIQ.
Kasli can be connected to the network using a 10000Base-X SFP module, installed into the SFP0 cage. Kasli-SoC features a built-in Ethernet port to use instead. If configured as a DRTIO satellite, both boards instead reserve SFP0 for the upstream DRTIO connection; remaining SFP cages are available for downstream connections. Equally, if used as a DRTIO master, all free SFP cages are available for downstream connections (i.e. all but SFP0 on Kasli, all four on Kasli-SoC). Kasli can be connected to the network using a 10000Base-X SFP module, installed into the SFP0 cage. Kasli-SoC features a built-in Ethernet port to use instead. If configured as a DRTIO satellite, both boards instead reserve SFP0 for the upstream DRTIO connection; remaining SFP cages are available for downstream connections. Equally, if used as a DRTIO master, all free SFP cages are available for downstream connections (i.e. all but SFP0 on Kasli, all four on Kasli-SoC).
The DRTIO line rate depends upon the RTIO clock frequency running, e.g., at 125MHz the line rate is 2.5Gbps, at 150MHz 3.0Gbps, etc. See below for information on RTIO clocks. The DRTIO line rate depends upon the RTIO clock frequency running, e.g., at 125MHz the line rate is 2.5Gbps, at 150MHz 3.0Gbps, etc. See below for information on RTIO clocks.
KC705 KC705 and ZC706
^^^^^
An alternative target board for the ARTIQ core device is the KC705 development board from Xilinx. It supports the NIST CLOCK and QC2 hardware (FMC).
Common problems
^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^
Two high-end evaluation kits are also supported as alternative ARTIQ core device target boards, respectively the Kintex7 `KC705 <https://www.xilinx.com/products/boards-and-kits/ek-k7-kc705-g.html>`_ and Zynq-SoC `ZC706 <https://www.xilinx.com/products/boards-and-kits/ek-z7-zc706-g.html>`_, both from Xilinx. ZC706, like Kasli-SoC, runs on the ARTIQ-Zynq port. Both are supported in several set variants, namely NIST CLOCK and QC2 (FMC), either available in master, satellite, or standalone variants. See also :doc:`building_developing` for more on system variants.
Common KC705 problems
"""""""""""""""""""""
* The SW13 switches on the board need to be set to 00001. * The SW13 switches on the board need to be set to 00001.
* When connected, the CLOCK adapter breaks the JTAG chain due to TDI not being connected to TDO on the FMC mezzanine. * When connected, the CLOCK adapter breaks the JTAG chain due to TDI not being connected to TDO on the FMC mezzanine.
* On some boards, the JTAG USB connector is not correctly soldered. * On some boards, the JTAG USB connector is not correctly soldered.
@ -111,11 +140,13 @@ VADJ
With the NIST CLOCK and QC2 adapters, for safe operation of the DDS buses (to prevent damage to the IO banks of the FPGA), the FMC VADJ rail of the KC705 should be changed to 3.3V. Plug the Texas Instruments USB-TO-GPIO PMBus adapter into the PMBus connector in the corner of the KC705 and use the Fusion Digital Power Designer software to configure (requires Windows). Write to chip number U55 (address 52), channel 4, which is the VADJ rail, to make it 3.3V instead of 2.5V. Power cycle the KC705 board to check that the startup voltage on the VADJ rail is now 3.3V. With the NIST CLOCK and QC2 adapters, for safe operation of the DDS buses (to prevent damage to the IO banks of the FPGA), the FMC VADJ rail of the KC705 should be changed to 3.3V. Plug the Texas Instruments USB-TO-GPIO PMBus adapter into the PMBus connector in the corner of the KC705 and use the Fusion Digital Power Designer software to configure (requires Windows). Write to chip number U55 (address 52), channel 4, which is the VADJ rail, to make it 3.3V instead of 2.5V. Power cycle the KC705 board to check that the startup voltage on the VADJ rail is now 3.3V.
Variant details
---------------
NIST CLOCK NIST CLOCK
^^^^^^^^^^ ^^^^^^^^^^
With the CLOCK hardware, the TTL lines are mapped as follows: With the KC705 CLOCK hardware, the TTL lines are mapped as follows:
+--------------------+-----------------------+--------------+ +--------------------+-----------------------+--------------+
| RTIO channel | TTL line | Capability | | RTIO channel | TTL line | Capability |
@ -155,11 +186,21 @@ The board has RTIO SPI buses mapped as follows:
The DDS bus is on channel 27. The DDS bus is on channel 27.
The ZC706 variant is identical except for the following differences:
- The SMA GPIO on channel 18 is replaced by an Input+Output capable PMOD1_0 line.
- Since there is no SDIO on the programmable logic side, channel 26 is instead occupied by an additional SPI:
+--------------+------------------+--------------+--------------+--------------+
| RTIO channel | CS_N | MOSI | MISO | CLK |
+==============+==================+==============+==============+==============+
| 26 | PMOD_SPI_CS_N | PMOD_SPI_MOSI| PMOD_SPI_MISO| PMOD_SPI_CLK |
+--------------+------------------+--------------+--------------+--------------+
NIST QC2 NIST QC2
^^^^^^^^ ^^^^^^^^
With the QC2 hardware, the TTL lines are mapped as follows: With the KC705 QC2 hardware, the TTL lines are mapped as follows:
+--------------------+-----------------------+--------------+ +--------------------+-----------------------+--------------+
| RTIO channel | TTL line | Capability | | RTIO channel | TTL line | Capability |
@ -193,38 +234,11 @@ The board has RTIO SPI buses mapped as follows:
There are two DDS buses on channels 50 (LPC, DDS0-DDS11) and 51 (HPC, DDS12-DDS23). There are two DDS buses on channels 50 (LPC, DDS0-DDS11) and 51 (HPC, DDS12-DDS23).
The QC2 hardware uses TCA6424A I2C I/O expanders to define the directions of its TTL buffers. There is one such expander per FMC card, and they are selected using the PCA9548 on the KC705. The QC2 hardware uses TCA6424A I2C I/O expanders to define the directions of its TTL buffers. There is one such expander per FMC card, and they are selected using the PCA9548 on the KC705.
To avoid I/O contention, the startup kernel should first program the TCA6424A expanders and then call ``output()`` on all ``TTLInOut`` channels that should be configured as outputs. To avoid I/O contention, the startup kernel should first program the TCA6424A expanders and then call ``output()`` on all ``TTLInOut`` channels that should be configured as outputs. See :mod:`artiq.coredevice.i2c` for more details.
See :mod:`artiq.coredevice.i2c` for more details. The ZC706 is identical except for the following differences:
.. _core-device-clocking: - The SMA GPIO is once again replaced with PMOD1_0.
- The first four TTLs also have edge counters, on channels 52, 53, 54, and 55.
Clocking
--------
The core device generates the RTIO clock using a PLL locked either to an internal crystal or to an external frequency reference. If choosing the latter, external reference must be provided (via front panel SMA input on Kasli boards). Valid configuration options include:
* ``int_100`` - internal crystal reference is used to synthesize a 100MHz RTIO clock,
* ``int_125`` - internal crystal reference is used to synthesize a 125MHz RTIO clock (default option),
* ``int_150`` - internal crystal reference is used to synthesize a 150MHz RTIO clock.
* ``ext0_synth0_10to125`` - external 10MHz reference clock used to synthesize a 125MHz RTIO clock,
* ``ext0_synth0_80to125`` - external 80MHz reference clock used to synthesize a 125MHz RTIO clock,
* ``ext0_synth0_100to125`` - external 100MHz reference clock used to synthesize a 125MHz RTIO clock,
* ``ext0_synth0_125to125`` - external 125MHz reference clock used to synthesize a 125MHz RTIO clock.
The selected option can be observed in the core device boot logs and accessed using ``artiq_coremgmt config`` with key ``rtio_clock``.
As of ARTIQ 8, it is now possible for Kasli and Kasli-SoC configurations to enable WRPLL -- a clock recovery method using `DDMTD <http://white-rabbit.web.cern.ch/documents/DDMTD_for_Sub-ns_Synchronization.pdf>`_ and Si549 oscillators -- both to lock the main RTIO clock and (in DRTIO configurations) to lock satellites to master. This is set by the ``enable_wrpll`` option in the :ref:`JSON description file <system-description>`. Because WRPLL requires slightly different gateware and firmware, it is necessary to re-flash devices to enable or disable it in extant systems. If you would like to obtain the firmware for a different WRPLL setting through AFWS, write to the helpdesk@ email.
If phase noise performance is the priority, it is recommended to use ``ext0_synth0_125to125`` over other ``ext0`` options, as this bypasses the (noisy) MMCM.
If not using WRPLL, PLL can also be bypassed entirely with the options
* ``ext0_bypass`` (input clock used directly)
* ``ext0_bypass_125`` (explicit alias)
* ``ext0_bypass_100`` (explicit alias)
Bypassing the PLL ensures the skews between input clock, downstream clock outputs, and RTIO clock are deterministic across reboots of the system. This is useful when phase determinism is required in situations where the reference clock fans out to other devices before reaching the master.

View File

@ -73,7 +73,7 @@ Core device management tool
.. automodule:: artiq.frontend.artiq_coremgmt .. automodule:: artiq.frontend.artiq_coremgmt
The core management utility gives remote access to the core device logs, the :ref:`core device flash storage <core-device-flash-storage>`, and other management functions. The core management utility gives remote access to the core device logs, the :ref:`core device flash storage <configuration-storage>`, and other management functions.
To use this tool, it is necessary to specify the IP address your core device can be contacted at. If no option is used, the utility will assume there is a file named ``device_db.py`` in the current directory containing the :ref:`device database <device-db>`; otherwise, a device database file can be provided with ``--device-db`` or an address directly with ``--device`` (see also below). To use this tool, it is necessary to specify the IP address your core device can be contacted at. If no option is used, the utility will assume there is a file named ``device_db.py`` in the current directory containing the :ref:`device database <device-db>`; otherwise, a device database file can be provided with ``--device-db`` or an address directly with ``--device`` (see also below).