Bidirectional Packet Protocol for FPGA communication
Open, Needs TriagePublic

Description

To apply for this task please complete this mandatory coding challenge together with your application: T871
Applications without matching coding challenge completed will not be considered.


The AXIOM Beta features two Lattice MachXO2 in addition to the Xilinx ZYNQ SoC which act as routing fabrics and extend the limited IOs from the main FPGA. The MachXO2s are connected via a single LVDS pair and share a common clock with the ZYNQ. A Packet Protocol is required to utilize the bandwidth and support various bus protocols on the Lattice FPGAs (I2C, SPI, GPIO ...)

Goals:

  • Define a bidirectional packet oriented protocol which works over a single LVDS pair.
  • Implement and test/simulate the packet oriented communication.
  • Implement various bus mappings utilizing the new interface (I2C, SPI, GPIO, ...).
  • Optimize communication with encoding and SERDES.

Prerequisites:

  • Packet Protocol Know-How
  • Encoding Know-How

Language Skills:

  • HDL (VHDL or Verilog)

Mentors: Bertl

To get in touch with any mentor check the Mentor Contact List.

Notes: If you do not have access to the required hardware platforms/dev kits, we will provide them (or remote access) to you for the duration of the project.

Related Objects

Bertl created this task.Feb 9 2017, 4:07 PM
Bertl updated the task description. (Show Details)
Bertl raised the priority of this task from to Needs Triage.
Bertl moved this task to FPGA / HDL on the Google Summer of Code 2018 board.
Bertl added a subscriber: Bertl.
Bertl updated the task description. (Show Details)Feb 9 2017, 5:23 PM

Hi.
So I read the description of this project, I assume that the bi-directional communication protocol to be implemented in both the ZYNQ and Machxo2 side. In case of packet-oriented communication, should we implement an architecture similar to the UDP protocol, which I think will be best for large bandwidth video data?

Also, the bus mapping part, does it involve synchronous communication between the zynq and MachXo2 without any time slack?

Bertl added a comment.Mar 12 2017, 8:17 PM

Bi-directional communication between ZYNQ and MachXO2 over a single LVDS pair (half duplex) in one case and over two LVDS pairs (full-duplex) in the other case.
Note that there is also a single ended clock line which can (and probably should :) be used for synchronization.

We do not want to transfer large bandwidth data like video streams (at least not for now :) over this connection, instead we want to reach devices connected to the routing fabrics (MachXO2) in an efficient and, at least for some devices, timely deterministic way (i.e. with a fixed, known delay).

The packet architecture should probably be more like ATM, i.e. fixed size, small packets which can be categorized into three priorities: realtime (immediate), efficient (prioritized) and idle.
The highest priority (realtime) will only be used to map 'wires' which basically function as triggers to e.g. the sensor or similar and require a deterministic delay.
The second priority (efficient) is basically for all devices connected to remote bus systems (I2C, SPI, etc) which need to be accessed via the routing fabrics.
The lowest priority (idle) is for general information exchange like communicating error counters or collected status information which is neither time critical nor tied to a device.

Hope that clarifies,
Herbert

danuka added a subscriber: danuka.Mar 23 2017, 4:25 PM
danuka added a comment.EditedMar 23 2017, 9:12 PM

Hi Bertl,
I am final year undergraduate of Electronic and Telecommunication Engineering, University of Moratuwa. I have gone through the description of the project that you are posted and I am clarifying on it. I have done many FPGA related projects and I am very interested in this field. I have few questions on this project.
• Do we have to implement the communication interface in the both Xilinx and lattice fpga platforms?
• If yes then will we be able to use the ip cores like xilincs fifos?
• Or else should it be a single implementation which can be run on both platforms?

Bertl added a comment.Mar 24 2017, 3:53 AM

The implementation has to be done on both FPGAs (Xilinx ZYNQ and Lattice MachXO2).
As this is a FOSS/OH project, you have to limit the design to IP without restrictions on both sides.
Using platform specific primitives like DDR I/O, SerDes or FIFOs is not as problem, as long as they are variants to a generic entity (which can be adapted to a new platform with little to none effort).

Hope that clarifies,
Herbert

Bertl updated the task description. (Show Details)Jan 23 2018, 12:44 PM
Bertl changed the edit policy from "All Users" to "Administrators".Jan 23 2018, 1:07 PM
sebastian updated the task description. (Show Details)Mar 8 2018, 8:24 PM
ymc98 added a subscriber: ymc98.Mar 22 2018, 12:03 AM

Balance of generality and simplicity is essential and I recommend glancing at PCIe TLP packages.
Possibly a much-simplified version of this would work here.

Are you bridging existing protocols, like, say AXI4 or Avalon?

The main focus will be on I2C, SPI and GPIO signals, where the latter ones should have predictable delays because they might control timing sensitive signals (exposure control, trigger, etc). There is no need to handle large bandwidth or complex communication like AMBA here.

Best,
Herbert