DØ Level 2 beta Firmware FPGAs etc Drew

DØ Level 2 beta Firmware, FPGAs, etc…. Drew Baden University of Maryland April 26, 2001 http: //macdrew. physics. umd. edu/cms/

DØ Level 2 beta Overview MBUS • Goal: Interface between MBUS and PCI • Implementation – Single large FPGA Xilinx XCV 405 -EM • 10, 800 logic cells • 573, 440 bits true dual-port block RAM Xilinx XVC 405 E – 71, 680 bytes, >4 k MBUS words • 404 I/O pins (not including load, JTAG, etc. ) BGA • Internal support for LVDS or PECL differential signaling @ 622 Mbps • 8 digital delay-locked loops (DLLs) per device – Eliminates clock skew for geographically separated modules PLX • DLLs can provide 2 x frequency multiplication • JTAG for boundary scans and “TAP” – Can use as an internal logic analyzer CPU DØ Level 2 ßeta 2

DØ Level 2 beta • PCI Implementation Overview (cont) – PLX 9054 PCI bus master interface • • 32 bit 33 MHz PCI 2 DMA enginers Programmable burst management Programmable interrupt generator 3. 3 V tolerant Asynchronous local bus/PCI bus Big Little endian conversion 132 MB/s theoretical bandwidth – How fast can it really go? • CMS group has built a PMC card with earlier version • They got ~110 MB/s DMA burst • We need around 80 MB/s • MBT can probably only send ~110 MB/s • Well matched DØ Level 2 ßeta 3

DØ Level 2 beta PLX 9054 • Local bus can run up to 50 MHz • Has some onboard FIFOs • DMA only as a local bus master – This is ok – we can send it a message as a local bus slave to initiate DMA – Can limit this to • DMA Address • GO • DMA size – This can also be eliminated if we use fixed burst size – 3 words x 25 ns (40 MHz) is 75 ns, very small overhead as long as DMA bursts are “large” • Ours are typically 10 -20 estimated DØ Level 2 ßeta 4

DØ Level 2 beta • 15 VME: FPGA I/O pins – 9 D 0 specific I/O, 6 for geographic addressing • 32 ECL scalers out • 193 MBUS: – 5 MBUS arbitration: • BOSS in, BOSS out, BOSSREQ, BOSSGRIN, BOSSGROUT – 4 MBUS handshaking • DSTROBE in, DSTROBE out, DDONE in, DDONE out – 33 MBUS address • 1 MBAD direction, 32 MBAD – 129 MBUS data • 1 MBDATA dir, 128 MBDATA – 14 MOD_DONE lines – 8 Miscellaneous • RD_WR, RD_WR_DIR, START_LOAD, DONE_OUT, MB_RESET, AP_FIFO_EMPTY, EV_LOADED(0) in and out DØ Level 2 ßeta 5

DØ Level 2 beta FPGA I/O pins (cont) • 19 Front panel – Clock, reset, 12 LEDs, 2 switches, 3 rotary switch • 32 Logic Analyzer test points • 89 Local Bus (aka “Add-on Bus”) Total Pin count: 380 Available on XCV 405 E: 404 • We will use them all, don’t worry! DØ Level 2 ßeta PLX M-mode bus J-mode is similar Add-on bus 89 signals CCS# LCLK LFRAME# LINT# LRESETo# USERi USERo BB# BDIP# BG# BIGEND# BR# DP[0: 3] LA[0: 31] LD[0: 31] MDREQ# RD/WR# RETRY# TA# TEA# TSIZ[0: 1] 6

DØ Level 2 beta Comments on 405 -EM • Current MBT firmware uses 10 k 100+10 k 50 – ~8000 logic cells total • 10, 800 for 405 -EM – 20% more resources – Currently 33% resource utilitization. • Will probably go up to 40%, call it 50% – Supports MBUS broadcast, MBUS PIO, FIFO reading, Cypress output, L 2 Global output, VME, 10 k 50 10 k 100, etc. • Functionality for Xilinx 405 -EM – MBUS side: • Data FIFO load (service MBUS broadcast) • PIO – PLX side: • FIFO read • PCI DMA transfer • PCI I/O to MBUS • Conclusion: – We do not expect this firmware to require significantly more resources than for MBT – it should fit ok with resources to spare DØ Level 2 ßeta 7

DØ Level 2 beta FPGA Contingency • XCV 812 E – >x 2 Logic Cells – x 2 block RAM – x 2 price! • This requires that we have a LOT of firmware simulated and synthesized this summer – This is well ahead of Sept schedule for boards – If it will not fit, we will know by Sept DØ Level 2 ßeta 8

DØ Level 2 beta Firmware • Basic firmware modules: ORSAY Maryland – “Local bus” side • ORSAY – MBUS side M L B O • UMD – Interface U C A • Joint specification • 1 week late May scheduled meeting in Paris to iron out DØ Level 2 ßeta S L B U S 9

DØ Level 2 beta Firmware Tools • HDL language – Both groups have experience in Verilog • Simulation – ORSAY uses Cadence – UMD uses Aldec • Synthesis – Agreement to both use Synplify DØ Level 2 ßeta 10

DØ Level 2 beta Manpower • Orsay – In-house experts in Verilog and Xilinx – Engineers • UMD – In-house expert in Verilog (me) – Engineers expert in FPGA schematic coding – $20 k in UMD budget for engineers to contribute to firmware • Hans Breden is taking Verilog class next week • UVA – Hirosky wants to learn! • We will all remain flexible and chip in where we can • We believe we have enough manpower to succeed DØ Level 2 ßeta 11

DØ Level 2 beta Conclusions • FPGA choice is sound • Some concern in number of resources – I like to have a lot of breathing room – <50% resource utilization appears to be a good upper limit – If we need more, we need to know “now” (summer) • PLX chip is excellent – Well tested and used in industry – Same for CERN CMS group (Attila Racz, on their web site) • Manpower appears to be sufficient • Firmware functionality lends itself to separation in development – Orsay does “Local bus” side – Maryland does “MBus” side DØ Level 2 ßeta 12