“A Configurable Radiation Tolerant Dual-Ported Static RAM macro, designed in a 0.25 µm CMOS technology for applications in the LHC environment.” 8th Workshop on Electronics for LHC Experiments 9-13 Sept. 2002, Colmar, France K. Kloukinas, G. Magazzu, A. Marchioro CERN EP division, 1211 Geneva 23, Switzerland Overview Motive of Work Description of the macro-cell design Experimental Results Conclusions Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 2 Motive of Work Several Front-End ASICs for the LHC detectors are using the CERN DSM Design Kit in 0.25 µm commercial CMOS technology. Many ASICs require the use of rather large memories in Readout Pipelines, Readout Buffers and FIFOs. CERN DSM Design Kit lacks design automation tools for generating customized SRAM blocks. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 3 Proposed Design Built an SRAM macro-cell that can be configured in terms of word counts and bit organization by means of simple floorplanning procedures. Initially designed for the needs of the “Kchip” Front-End ASIC used in the CMS ECAL Preshower detector. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 4 CERN-SRAM specifications Data In Scalable Design Configurable Bit organization (n x 9-bit). Configurable Memory Size (128 – 4Kwords). Write Address Synchronous Dual-Port Operation Permits Read/Write operations on the same clock cycle. Typical Operating Frequency: 40 MHz. Full Static Operation. Divided Wordline Decoding. Radiation Tolerant Design KLOUKINAS Kostas SRAM SRAM RD WR CLK Low Power Design Sept 12, 2002 Read Address Data Out EP/CME-PS 5 Memory Cell RBL WBL WBL RBL BL BL WL R_WL W_WL Dual Port SRAM Cell Single Port SRAM Cell To minimize the macro-cell area a Single Port memory cell is used based on a conventional cross-coupled inverter scheme. Gain in Memory Cell Layout Area = 18% Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 6 Memory Cell Design BL GND BL VDD GND BL M1 BL M2 M3 8.42 µm WL WL PC Single Port memory cell Interconnect: 3 metal layers • 1st for local interconnects • 2nd for vertical bitlines and power lines • 3rd for horizontal wordlines Memory Cell Area: 47.152 µm2 Sept 12, 2002 KLOUKINAS Kostas 5.60 µm EP/CME-PS 7 Write Drivers Sept 12, 2002 Word-line Buffers SRAM Array • Registered Inputs • Latched Outputs n-7 clk Clk R W Dual-port functionality is realized with a time sharing access mechanism. Timing Logic Read Logic Column-decoder KLOUKINAS Kostas EP/CME-PS Data Latch RA <n-1..0> Addr. Reg. 7 Row decoder WA <n-1..0> m Addr. Reg. Din <m-1..0> Data Reg. SRAM Block Diagram Dout <m-1..0> 8 SRAM Interface Timing WRITE tS 1 tH READ tS 2 tH READ/WRITE 3 4 5 Clk WA RA W R Din tacc Dout Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 9 SRAM macro-cell Design Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 10 Address Mux Register Leaf cell is based on the D-F/F and the 2-input Mux standard cells found in the CERN DSM Design Kit. WA D SET CLR True & Complementary output with balanced timing. Easily sizeable by abutting the necessary number of leaf cells. RA D SET CLR Q Q Addr AddrZ Q Q Clk Leaf Cell Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 11 Row Decoder WL0 ¾Decoder: 7 to 128 ¾Hardwire-configured. ¾Pre-routed layout block. A6 WL1 WL2 ¾Dynamic NAND-type. ¾Speed, Area, Power advantages over the static NAND-type. ¾Latched output. WL3 A0 precharge WL evaluate WL128 A0...A6 A0...A6 Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 12 Column Decoder Static NAND-type implementation Column decoding is one of the last actions to be performed in the read sequence. It can be executed in parallel with other functions, and can be performed as soon as address is available. Its propagation delay does not add to the overall memory access time. Size Configurable Make use of Design kit standard cells. Decoding function is via-hole programmable. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 13 Divided Wordline Decoding Global Word-line Local Word-line Local Word-line Block Select Block Select Reduced Power Consumption. The non accessed portions of the memory remain in the precharge state. Improved Wordline Selection Time. Since the RC delay in each divided wordline is small due to its short length. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 14 Divided Wordline Decoding Data In Write Drivers Write Drivers Write Drivers Write Drivers Global Word-line SRAM Array Word-line Buffers Word-line Buffers Row Decoder SRAM Array Word-line Buffers SRAM Array Word-line Buffers Local Word-line SRAM Array Address Read Logic Read Logic Read Logic Read Logic Column Dec. Column Dec. Column Dec. Column Dec. Data Out Block Pre-Dec. Block Select signals Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 15 SRAM macro-cell Design Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 16 Data Input Output Ports Data Input Register Leaf cell is based on the D-F/F standard cell from CERN DSM Design Kit. True & Complementary output with balanced timing. Data Output Latch Leaf cell is based on the Latch standard cell from CERN DSM Design Kit. Easily sizeable by abutting the necessary number of leaf cells. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 17 SRAM Data Path Data in Clk Write Drivers Write enable Bit Line Bit Line BLPC precharge evaluate Word Line precharge evaluate Read Logic Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS Latch Data out 18 Read Logic Substitution of the conventional sense amplifier with an asymmetric inverter. Reduced Power Consumption Stable operation al low power supply voltages. Acceptable performance for target applications. Easy to design. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 19 Replica Techniques Dummy Bit Lines Wordline select time depends on the size of the memory. Dummy Wordline with replica memory cells to track the wordline charge-discharge time. Bitline Timing 128 rows Scalability Dummy Bitlines to mimic the delay of the bitline path over all conditions. SRAM Array WLdummy BL0 Dummy Word Line Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 20 Replica Techniques Data in Dummy Bit Lines WEN Row Decoder BLPC Bit Line Bit Line A6 Global Word Line A0 Local Word Line WLPC Block Select WLPC WLdummy LWLdummy Dummy Word Line BL Latch Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS BL BL0 Data out 21 Timing Logic Data Input Register Address Mux Register Clk SRAM Interface Clk Clk R W WLpc BLpc Timing Logic Memory Cell WLdummy BL0 Array Sept 12, 2002 KLOUKINAS Kostas Memory Cell REN Array WEN Latch EP/CME-PS Data Ouput Latch 22 Timing Logic Asynchronous internal timing of control signals. Static operation. Hand-shaking and transition detection to realize internal timing loops. Timing loops are initiated by the system clock and terminated upon completion of the operation. All control signals are forced back to their initial state to prepare for subsequent tasks. During standby periods, bitlines and wordlines precharge-evaluate cycles are not initiated, thus keeping the Power Consumption to a minimum. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 23 Operation and Timing Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 24 Cell Library Fixed Layout Size Configurable Data Input Register WordLine Buffers SRAM column, 128 x 9bits (50.4 µm x 1086.2 µm) Column Decoder Block Pre-Decoder Output Data Latche Row Decoder Address Mux Register Timing logic Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 25 Floorplanning Column Timing PreDecoder Address Reg Block Read logic Read logic Read logic Read logic Column Decoder DOut Latch Read logic Read logic Read logic Read logic DIn Reg Write drv. Write drv. Write drv. Write drv. Row Decoder (128 rows) Write drv. Write drv. Write drv. Write drv. vdd gnd Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 26 DIn WA RA CLK W R DOut CAD Tools Support Digital Simulation Digital Simulator SRAM verilog module (parameterized) Sept 12, 2002 VERILOG KLOUKINAS Kostas EP/CME-PS 27 CAD Tools Support Logic Synthesis Template SRAM timing Design Kit .lib file compilation Combined .lib file Combined .db file Logic Synthesis Tool SYNOPSYS Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 28 CAD Tools Support Place & Route Template SRAM timing Design Kit TLF file Layout view Sept 12, 2002 compilation Combined TLF file Abstract view Combined CTLF file LEF file KLOUKINAS Kostas Place & Route Tool Silicon Ensemble EP/CME-PS 29 Experimental Results To prove the concept of the SRAM macro-cell scalability and to evaluate the performance of the proposed design we have fabricated two test chips: a 1Kwords X 9bits and a 4Kwords X 9bits. Both chips were tested and found functional. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 30 Submitted SRAM Chips 1st Prototype Design: CERN_SRAM_1K Configuration: 1K x 9 bit Size: ~560µm x 1,300µm Area: ~0.73mm2 Density: ~12.6Kbit/mm2 The Memory consists of 2 Blocks of 512 x 9bits. Each Block is composed by 4 Columns of 128 X 9bits. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 31 Submitted SRAM Chips 2nd Prototype Design: CERN_SRAM_4K Configuration: 4K x 9 bit Size: ~1,850µm x 1,300µm Area: ~2.4mm2 Density: ~15.4Kbit/mm2 The Memory consists of 8 Blocks of 512 x 9bits. Each Block is composed by 4 Columns of 128 X 9bits. Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 32 CERN SRAM test results Test chip: 4Kx9bit Functional tests Max operating frequency: Simultaneous Read/Write operations: 70MHz @ 2.5V Read access time: 7.5ns @ 2.5V Power dissipation: 15µW / MHz @ 2.5V for simultaneous Read/Write operations on the same clock cycle (0.60mW @ 40MHz). Tests for process variations: Differences in the access time < 1ns for: -3σ, –1.5σ, typ, +1.5σ, +3σ Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 33 Performance Tests Test Chip: 4Kword X 9bits Operation Frequency: 50MHz Power Supply: 2.5Volts Read Access Time: 7.5nsec Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 34 Performance Tests Schmoo Plot All 1’s and all 0’s Checkerboard Marching 1’s Marching 0’s 2.7 2.6 Power Supply Voltage (V) Test Chip: 4Kword X 9bits Power Supply: 2.0 - 2.7Volts Operation Frequency: 50MHz Test Patterns: 2.5 Pass 2.4 2.3 2.2 2.1 2 1.9 6.00 8.00 10.00 12.00 14.00 16.00 Access Time (nsec) Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 35 Power dissipation Power dissipation of macro-cell. CERN SRAM 4K 1200.0 Test chip: 4Kwords x 9bits Power Dissipation (uW) 1000.0 800.0 Standby Idle 600.0 Read Write 400.0 Read/Write Power Operation (µW/MHz) Standby 0.10 Idle 1.90 Read 7.40 Write 10.60 Read/Write 14.05 200.0 0.0 20 30 40 50 60 70 Clock frequency (MHz) Test Conditions Operation Standby Idle Read Write Read/Write Sept 12, 2002 Description No operation, addr. & data static. No operation, addr. & data changing in every clk cycle checkerboard data pattern checkerboard data pattern checkerboard data pattern KLOUKINAS Kostas EP/CME-PS 36 Irradiation Tests Ionizing Total Dose Test chip: 4Kwords x 9bit Conditions Source: X-rays. Step Irradiation: 1Mrad, 5Mrad, 10Mrad. Constant dose rate: 21.2 Krad/min. Annealing: 24h @ ~25 oC. Under bias, in Standby mode during irradiation & annealing. Results No increase in power dissipation. No measurable degradation in performance. Single Event Upset: Under preparation Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 37 CERN SRAM popularity ! z ATLAS MCC chip z z z z z z z z z z z Sept 12, 2002 z z KLOUKINAS Kostas Memory configuration: 128 x 153 bits Detector: ATLAS TRT Lab: CERN CMS Kchip z z z Memory configuration: 256 X 9 bits Detector: LHCb muon system Lab: INFN Cagliary Memory configuration: 128 x 18bit Detector: ATLAS tracker Lab: NEVIS Labs ATLAS DTMROC chip z z LHCb SYNC chip z z Memory configuration: 16K X 9 bits Detector: ALICE Silicon Drift Det. Lab: INFN Torino Memory configuration: 256 X 9 bits Detector: ALICE Silicon Drift Det. Lab: INFN Bologna ATLAS SCAC chip z z ALICE CARLOS chip z z Memory configuration: 128 x 27bit Detector: ATLAS PIXEL Lab: INFN Genova ALICE AMBRA chip z z z Memory configuration: 2K x 18 bits 128 x 18 bits Detector: CMS Preshower Lab: CERN Chips submitted and tested EP/CME-PS 38 Design Support Delivery Delivery of of SRAM SRAM design design library library Half Half aa day day “design “design course” course” @ @ CERN CERN Designer Designer configures configures his his macrocell macrocell Review Review the the macrocell macrocell design design Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 39 Conclusions Design Status Design meets target specifications. Macrocell has been successfully used in a number of ASIC designs. Future Plans No further development is foreseen. Design Support Contact Person: Kostas.Kloukinas@CERN.ch Information on the Web http://home.cern.ch/kkloukin Sept 12, 2002 KLOUKINAS Kostas EP/CME-PS 40 vdd gnd Sept 12, 2002 CLK W R DOut Read logic WorldLine Buffers & Block Decoders KLOUKINAS Kostas Read logic Column Decoder. EP/CME-PS Timing PreDecoder Address Reg RA Block Column WA DIn Reg Write drv. DOut Latch Read logic DOut Latch Timing Column Address Reg Block Write drv. DIn Row Decoder (128 rows) Row Decoder (128 rows) Write drv. DIn Reg Floorplanning DIn WA RA CLK W R DOut vdd gnd 41