The Road Warrior: first use of the Pulsar for SVT Elena Pedreschi, Marco Piendibene, Franco Spinella The problem: 5/5 + XFT To raise efficiency SVT can run in 4/5 mode + XFT. But … the tracks with all the 5/5 hits come outs from the AM 6 times … The TF receives each track and process it -> replicate tracks are processed 6 times -> this takes to much time to make 4/5 mode usable One possible solution is to flag each replicated track, so that the TF can skip it. The Road Warrior firmware, implemented in the Pulsar, performs this function
How it is done ? We implement a pseudo-CAM (or a simple AM) : I LAYER XFT II LAYER III LAYER IV LAYER V LAYER ROAD PHI ROAD-PACKET: I LAYERII LAYERIII LAYERIV LAYERV LAYERXFT I LAYERII LAYERIII LAYERIV LAYERV LAYERXFT I LAYERII LAYERIII LAYERIV LAYERV LAYERXFT I LAYERII LAYERIII LAYERIV LAYERV LAYERXFT I LAYERII LAYERIII LAYERIV LAYERV LAYERXFT IF MATCH = 1 : PHI = 0xF ELSE STORE TMP_WORD TO NEXT AVAILABLE AM PATTERN OUT TO TF STORE EACH HIT IN A TMP REGISTER, IN THE RIGHT POSITION MATCH IF : AT LEAST 4/5 HITS + XFT 1 0
The logic is implemented in the Pulsar control FPGA The CAM is 64 words x 108 bits (64 is the maximum number roads/event) It is organized as a 3-stage pipeline The clock runs at 30 Mhz Works without adding any delay, except 3 clocks latency Implements a VME interface, copied from Muon pulsar firmware (thx Sakari) Through VME it is possible to choose a pass-through mode or rw mode Only 13 % of the control FPGA is used Testbench: we use an SVT merger as data generator and receiver. We use some standard SVT code (svtvme) to run the test The behavioural firmware has been simulated through Modelsim-Linux + Synopsys smart-model (for the FIFO model) Test: started 2 days ago, thanks to Ted help. The firmware seems working except some minor details
PLANS: We hope to finalize the version 1 test next week Some modifications to the firmware are foreseen: test mode, spy buffers etc (version2 ) Early next year we plan to install 12 pulsars in the SVT crates (one for each silicon wedge) A road warrior board could be inserted immediately after the AM: big SRAM required The HB + AMS + RW boards could be merged together to one single pulsar. Again a big SRAM is required (and a big amount of VHDL work … )