Session 20 – Analog Techniques

Oak Ballroom, Wednesday Morning, September 19

 

Chair:  Don Thelen

Co-Chair :  Dennis Fischette

 

This session presents analog circuit design techniques used in frequency selective filters, image sensors, and temperature and process compensated oscillators.

 

8:25 am

Introduction

 

20.1

8:30 am

An 80MHz Noise Optimized Continuous-Time Bandpass Filter in 0.25 µm  BiCMOS,  A. Kumar and P. Allen, Georgia Institute of Technolgy

 

An 80 MHz bandpass filter with a tunable quality factor of 16~44 using an improved transconductor circuit is presented. The noise optimized biquad structure achieves SNR of 45dB at IMD of 40dB. The P-1dB compression point and IIP3 of the filter are -10dBm and -2.68dBm, respectively. The proposed filter gives an output of 1.87Vpp at -1dB compression point in 3.3V supply and consumes 66mW. It is fabricated in 0.25 µm  BiCMOS.

 

20.2

8:55 am

A Low Power 44-300 MHz Programmable Active-RC Filter in 0.18µm CMOS,  T. Laxminidhi, V. Prasadu and S. Pavan, Indian Institute of Technology-Madras

 

We present a widely programmable fifth order Chebyshev opamp-RC ladder filter  whose 3 dB bandwidth is digitally tunable over a 7X range, from  44-300 MHz.  The opamp uses feed-forward compensation to achieve high DC gain and wide bandwidth with reduced bias currents. The principle of ``constant capacitance scaling" is used  to maintain the shape of the filter response when the bandwidth is changed by a large factor. The filter consumes 36 mW from a 1.8V supply and has a dynamic range of 56.6 dB.

 

20.3

9:20 am

A Q-enhanced Transformer Coupling Dynamic Dual-Mode 5GHz Bandpass NB / Interference Rejection UWB Filter,  B. Pham and A. Dinh, University of Saskatchewan

 

This paper presents a fully integrated dual mode Q-enhanced bandpass/notch filter front-end designed for Narrowband and UWB receivers. Using a simple logic state to switch its function, the filter can provide bandpass filtering for channel selection in the NB mode or NB interference rejection in the UWB mode. A Q-enhanced circuit coupled through a transformer is used to realize the modes. Using 1.8V supply, the filter consumes 22 mW of power and provides a 32dB interference rejection in the UWB mode and a gain of 24dB in the NB mode.  The design occupies a die area of 1mm2 using the mainstream 0.18µm CMOS process.

 

20.4

9:45 am

A Process and Temperature Compensated Two-Stage Ring Oscillator,  K. Lakshmikumar, V. Mukundagiri and S. Gierkink, Conexant Systems

 

Local positive feedback in a delay element enables a ring oscillator with only two stages to oscillate and produce quadrature clocks. Routh-Hurwitz’s criterion is applied to prove that such a structure can oscillate. An internally generated power supply from a constant-gm bias keeps the free running frequency to within ±5% from -40 to 125°C over process variations.  The 1.25GHz oscillator in 0.13 µm  CMOS draws 3.4mA and has a phase noise of -88dBc/Hz at 1MHz offset.

 

10:10 am - Break

 

20.5

10:25 am

Signal Processing Architectures for Low-Noise High-Resolution CMOS Image Sensors (INVITED PAPER),  S. Kawahito, Shizuoka University

 

In this paper, signal processing architectures for low-noise high-resolution CMOS image sensors are reviewed and discussed.  For low-noise signal readout, a column amplifier plays an important role for reducing both the noises due to a wideband output buffer and a pixel source follower. With high amplifier gain, a double-stage noise canceling technique and an advanced signal processing using oversampling techniques effectively reduces the noise due to the pixel source follower. Architectures and topologies for on-chip A/D conversion including pixel parallel, column parallel and serial schemes are discussed. On-chip column parallel analog-to-digital (A/D) conversion is particularly important for low-noise and high-speed signal readout.

 

20.6

10:50 am

A 1.8 mm2, 11 mA, 23.2 dB-NF, discrete-time filter for GSM/WCDMA/WLAN using retiming technique,  T. Sano, T. Maruyama, I. Yasui, H. Sato and T. Shimizu, Renesas Technology

 

The discrete-time filtering is suitable for pass band selection and anti-alias filtering. Reduction of large switched-capacitor area is a key issue for the filter. In this paper, we propose the retiming technique and the optimization of unit capacitor to reduce the number of capacitors. The area of the filters is 1.8 mm2. Noise Figure (NF) and current consumption in GSM mode are 23.2 dB and 11 mA, respectively.

 

20.7

11:15 am

An Equalized Ultra-Wideband Channel-Select Filter with a Discrete-Time Charge-Domain Band-Pass IIR Filter,  A. Yoshizawa and S. Iida, Sony Corporation

 

An ultra-wideband channel-select filter with equalized pass-band characteristics has been implemented with a 0.13 µm  CMOS technology. A discrete-time charge-domain IIR band-pass filter is used to compensate degradation of the 3rd-order continuous-time LC-LPF and the sinc response of the 1st-order charge-domain LPF. Measurements show that the IIR BPF reshapes the pass-band, allowing a 3-dB bandwidth of 500 MHz with a notch at 1 GHz. The filter dissipates 18.8 mW from a 1.2-V supply voltage.

 

 

 

Session 21 – High Performance Processors

and Digital Techniques

Fir Ballroom, Wednesday Morning, September 19

 

Chair:  Aurangzeb Khan

Co-Chair:  Henry Chang

 

This session examines processor architectures, design methods, and implementations in 90nm and 65nm.  Additional digital techniques are explored in the areas of low power, reliability and emulation-based verification.

 

8:25 am

Introduction

 

21.1

8:30 am

Cell Broadband Engine Processor Design Methodology (INVITED PAPER),  O. Takahashi, E. Behnen, S. Cottier, P. Coulman, S. Dhong, B. Flachs, P. Hofstee, C. Johnson and S. Posluszny, IBM

 

The Cell BE design methodology is described which enables high frequency, high performance, power efficient, and area optimized design. It includes a hierarchical design style, clean clock boundary, effective use of non-scan latches, at-speed scan testing, custom design like synthesized macro, fine grained clock gating scheme, and cycle accurate power analysis.

 

21.2

9:20 am

Implementation of the 65nm Cell Broadband Engine,  M. Riley, B. Flachs, S. Dhong, G. Gervais, S. Weitzel, M. Wang, D. Boerstler, M. Bolliger, J. Keaty, J. Pille, R. Berry,and O. Takahashi, IBM, Y. Nishino, Sony Computer Entertainment, T. Uchino, Toshiba America

 

The first generation Cell Broadband Engine processor introduced the CELL Architecture that consists of nine processor cores fabricated in the 90nm CMOS SOI technology.  This paper describes the advances made by moving Cell Broadband Engine design from 90nm CMOS SOI to 65nm CMOS SOI.

 

21.3

9:45 am

Compact Fault Recovering Flip-Flop with Adjusting Clock Timing Triggered by Error Detection,  S. Yasuda and S. Fujita, Toshiba Corporation

 

We propose novel compact fault recovering flip-flops (CFR-FFs) which can recover timing error and soft error. Two kinds of CFR-FFs are proposed. One is that the clock rises every cycle, and the other is that the clock rises only when data has changed. Error rates were experimentally measured by applying a noise signal to the supply voltage. It is confirmed that the proposed CFR-FFs can highly suppress the errors more than a conventional FF.

 

10:10 am - Break

 

21.4

10:25 am

A 2GHz, 7W (max) 64b Power™ Microprocessor Core,  D. Murray, J. Burnette, B. Campbell, M. Chung, B. Fernandes, S. Ghosh, R. Goel, G. Hess, H. Huang, Z. Huang, N. Javarappa, P. Kanapathipilai, F. Klass, F. Liu, A. Mehta, Y. Modukuru, N. Nerurkar, A. Radhakrishnan, S. Santhanam, J. Sugisawa, S. Sundar,  H. J. Tam, R. Wen E. Wu, J.-C. Yeh, J. Yong, and Z. Zambare, PA Semi, Inc.

 

The PA6T core is an out-of-order superscalar implementation of the Power Architecture. Power efficiency is achieved through micro-architecture, logic, and circuit optimizations. The processor is fabricated in a 65nm, triple Vt, dual oxide 8M CMOS process. Worst-case power dissipation at 2GHz is 7W.

 

21.5

10:50 am

Power-Efficient Dual-Supply 64kB L1 Caches in a 65nm CMOS Technology,  B. Campbell, J. Burnette, N. Javarappa and V. von Kaenel, PA Semi, Inc.

 

The 64kB L1 caches on the PA6T-1682M SoC CPU are composed of common data and tag structures fabricated in a 65nm CMOS process and deliver a 1.5 cycle read latency with 32GB/s bandwidth at 2GHz.  Several features optimize cache performance and power including power-down safe level shifters, streamlined dual-supply bitslices, fine-grain clock gating, and a centralized tag floorplan.

 

21.6

11:15 am

Process-Tolerant Low-Power Adaptive Pipeline under Scaled-Vdd,  S. Ghosh, P. Batra, K. Kim and K. Roy, Purdue University

 

We implement a low-power and robust pipeline design methodology suitable for aggressive voltage scaling while maintaining high frequency operations. We isolate the critical paths; make them predictable (by design) and ensure they are activated rarely. At scaled supply (frequency unchanged), possible delay errors (under 1-cycle) are predicted ahead in time and avoided by adaptively stretching the clock period. Test-chip implementation in 130nm process shows 40% power savings with 13% performance loss and ~9.4% area overhead.

 

21.7

11:40 am

ASIC Design and Verification in an FPGA Environment,  D. Markovic, C. Chang, B. Richards, H. So, B. Nikolic and R. Brodersen, University of California Berkeley

 

A unified algorithm-architecture-circuit co-design environment for dedicated signal processing hardware is presented.  The approach is based on a single design description in the graphical Matlab/Simulink environment that is used for FPGA emulation, ASIC design, verification and chip testing.  The tool flow propagates up circuit-level performance and power estimates to rapidly evaluate architecture-level tradeoffs.  The common Simulink design description minimizes errors in translation of the design between different descriptions, and eases the verification burden.  The approach is demonstrated on an ASIC for 4x4 MIMO signal processing.

 

 

 

Session 22 – mm-Wave Systems and Building-Blocks

Pine Ballroom, Wednesday Morning, September 19

 

Chair:  Cicero Vaucher

Co-Chair:  John Rogers

 

This session will present advances in mm-Wave transmitters and receivers.  In addition the design of on-chip programmable phase shifters and mm-wave power amplifiers will be addressed.

 

8:25 am

Introduction

 

22.1

8:30 am

mm-Wave Silicon ICs: Challenges and Opportunities (INVITED PAPER),  A. Hajimiri, California Institute of Technolgy

 

Millimeter-waves offer promising opportunities and interesting challenges to silicon integrated circuit and system designers. These challenges go beyond standard circuit design questions and span a broader range of topics including wave propagation, antenna design, and communication channel capacity limits. It is only meaningful to evaluate the benefits and shortcoming of silicon-based mm-wave integrated circuits in this broader context. This paper reviews some of these issues and presents several solutions to them.

 

22.2

9:20 am

65-nm CMOS, W-Band Receivers for Imaging Applications,  K. Tang, M. Khanpour, S. Voinigescu, University of Toronto, P. Garcia, C. Garnier, STMicroelectronics

 

Two 76-92 GHz receivers, featuring 3-stage cascode LNAs coupled through a transformer to a double-balanced Gilbert-cell mixer and differential DC-5GHz IF buffer, are reported in 65-nm general purpose (GP) CMOS technology.  One receiver features a traditional LNA with series-series inductive feedback, while the LNA of the second receiver employs a shunt-series, transformer-feedback cascode stage.  Both receivers have a differential down-conversion gain of 12 dB, an input P1dB of -13 dBm, and a double-sideband noise figure of 9-10 dB. They each occupy an area of 550um×550um and consume 94 mW. An LO-to-RF isolation of 60 to 59 dB was measured for LO signals in the 80-85 GHz range.  The transformer-feedback provides a broader bandwidth input match, lower than -10 dB from 74 to 95 GHz.

 

22.3

9:45 am

A 4-Channel 24-27 GHz UWB Phased Array Transmitter in 0.13µm CMOS for Vehicular Radar,  H. Krishnaswamy and H. Hashemi, University of Southern California

 

An Ultra-wideband (UWB) Variable-Phase Ring Oscillator (VPRO) and Phase-Locked Loop (PLL) architecture for integrated phased-array transmitters is presented.  A wideband VPRO operates at half the desired frequency range, and a squarer is used in each channel to cover the ±180o phase-shift range completely. A waveform-adaptive, tunable-narrowband design paradigm is introduced that greatly simplifies the design of the UWB blocks. A fully-integrated, 4-channel, 24-27 GHz, phased-array transmitter is implemented in 0.13µm CMOS to validate these claims.

 

10:10 am – Break

 

22.4

10:25 am

A CMOS 22-29GHz Receiver Front-End for UWB Automotive Pulse-Radars,  V. Jain, P. Heydari, University of California Irvine and S. Sundararaman, Avago Technologies

 

The design of a CMOS 22-29GHz pulse-radar receiver (RX) front-end for ultra-wideband (UWB) automotive radar sensors is presented. Fabricated in a 0.18µm CMOS process, the 3mm2 RX chip achieves a conversion gain of 35-38.1dB, noise figure of 5.5-7.4dB and input return loss less than -14.5dB in the 22-29GHz band. The phase noise of the constituent QVCO is -107dBc/Hz at 1MHz offset from a center frequency of 26.5GHz. The total dc power dissipation of the RX including LO/output buffers is 131mW.

 

22.5

10:50 am

An X- and Ku-Band 8-Element Linear Phased Array Receiver,  K.-J. Koh and G. M. Rebeiz, University of California, San Diego

 

This paper presents an 8-element linear phased array receiver adopting RF phase shifting architecture in 0.18- µm  SiGe BiCMOS technology for X- and Ku-band applications. The receiver shows about 20 dB of power gain per channel at 12 GHz with a 3-dB gain bandwidth from 8.5 to 14.5 GHz. The rms gain error is less than 0.9 dB and the rms phase error is less than 6 degree at 6-18 GHz for all the 4-bit phase states. The minimum NF is 3.85 dB at 10-11 GHz and typical input P1dB at 12 GHz is -31 dBm with 170 mA of current consumption from a 3.3 V supply voltage. The overall chip size is 2.2×2.45 mm2.

 

22.6

11:15 am

A 30-40 GHz 1:16 Internally Matched SiGe Active Power Divider for Phased Array Transmitters,  J. May and G. Rebeiz, University of California, San Diego

 

An active 1:16 30-40 GHz power divider was implemented in 0.18 µm  SiGe BiCMOS.  The 2 x 2 mm2 splitter exhibits 4.5 ± 1.5 dB total power gain with rms phase imbalance of less than 6º from 30 to 40 GHz across all 16 channels, consumes 190 mA from a 3.3 V supply, is 15x smaller than an equivalent Teflon-based PCB transmission-line 1:16 power divider, and is the first 1:16 Ka-Band SiGe active power divider.

 

22.7

11:40 am

A 60 GHz Power Amplifier in 90nm CMOS Technology,  B. Heydari, M. Bohsali, E. Adabi and A. Niknejad, University of California Berkeley

 

A two-stage 60 GHz 90nm CMOS PA has been designed and fabricated. The amplifier has a measured power gain of 9.8 dB. The input is gain matched while the output is matched to maximize the output power. The measured P-1dB =6.7 dBm with a corresponding power added efficiency of 20%.This amplifier can be used as a pre-driver or as the main PA for short range wireless communication. The output power can be boosted with on-chip or spatial power combining.

 

 

 

Session 23 – Methodology and Design for Process

Variability Mitigation

Cedar Ballroom, Wednesday Morning, September 19

 

Chair:  Philippe Jansen

Co-Chair:  Hamid Mahmoodi

 

This session addresses different aspects and approaches to mitigate process variability for yield enhancement for nano-scale CMOS technologies.  The discussed topics include yield estimation, design/process interaction, and design for yield.

 

8:25 am

Introduction

 

23.1

8:30 am

At Tape-out:  Can System Yield in Terms of Timing/Energy Specifications Be Predicted? (INVITED PAPER),  A. Papanikolaou, M. Miranda, P. Marchal, B. Dierickx and F. Catthoor, IMEC

 

Process variability is introducing uncertainty in all the system level parametric specifications. Existing variability aware techniques can only capture and model the variations on system timing and leakage power. This paper proposes a framework that can capture variability in the dynamic energy consumption as well. It percolates variability information from semiconductor process to the Register Transfer Level. This enables to capture the application dynamics and provide an accurate estimation of dynamic energy along with leakage and timing.

 

23.2

8:55 am

Process/Product Interactions in a Concurrent Design Environment (INVITED PAPER),  L. Bair, Advanced Micro Devices

 

The interactions between VLSI processes and the products built in them continue to perplex those who design and those who manufacture semiconductor chips.  Predicting, preventing, and minimizing these interactions is compounded by attempts to minimize time-to-market through concurrent process and design development in integrated design and manufacturing environments.  Past experience, engineering conservatism, and flexible design techniques enable successful concurrent deep submicron CMOS VLSI designs.

 

23.3

9:20 am

Dynamic Supply Noise Measurement with All Digital Gated Oscillator for Evaluating Decoupling Capacitance Effect,  Y. Ogasahara, M. Hashimoto and T. Onoye, Osaka University

 

This paper proposes an all digital measurement circuit called "gated oscillator" for capturing waveforms of dynamic power supply noise. The gated oscillator is constructed with standard cells, and thus easily embedded in SoCs. The performance of the gated oscillator is testified with fabricated test chips in a 90nm process. Characteristics of decoupling capacitance are discussed focusing on channel length and distance, based on supply noise waveforms measured by the gated oscillator.

 

23.4

9:45 am

An Embedded 8-bit RISC Controller for Yield Enhancement of the 90-nm PRAM, H. Kim, J. Yoo, H.-J. Yoo, KAIST, and K. Sohn, Samsung

 

An embedded 8b RISC for advanced memories is designed to control, analyze and optimize the memory timing and voltage parameters. The processor-based built-in-selfoptimize(BISO) algorithm is proposed to enhance the memory yield.  A test PRAM with the RISC is fabricated in 90nm, 3-metal diode-switch process.  By applying BISO, the PRAM margin window increases by 221%. It operates at 100MHz and consumes 28.4mW at 1.0V supply voltage. The embedded RISC enables 100Mb/s/pin read/write throughputs to PRAM.

 

10:10 am – Break

 

 

 

Session 24 – Advanced Memories

Cedar Ballroom, Wednesday Morning, September 19

 

Chair :  Kenji Noda

Co-Chair:  Sreedhar Natarajan

 

This session covers SRAM alternatives, and describes the evolution of electronically-programmed fuses to enable redundancy.

 

10:20 am

Introduction

 

24.1

10:25 am

A 180 Kbit Embeddable MRAM Memory Module,  J. J. Nahas, T. Andre, B. Garni, C. Subramanian, H. Lin, S.M. Alam, K. Papworth and W. L. Martino, Jr., Freescale Semiconductor

 

A 180 Kbit Magnetoresistive Random Access Memory (MRAM) designed for embedding in a 0.28 micron CMOS process has been developed. The memory cell is based on a 1-Transistor 1-Magnetic Tunnel Junction (1T1MTJ) bit cell. The architecture, write driver, and sense amplifier are described. The use of a Test Register to characterize and optimize the memory design is also discussed.

 

24.2

10:50 am

A 1.0GHz multi-banked embedded DRAM in 65nm CMOS featuring concurrent refresh and hierarchical BIST,  D. Anand, J. Covino, J. Dreibelbis, J. Fifield, K. Gorman, M. Jacunski, J. Paparelli, G. Pomichter, D. Pontius, M. Roberge and S. Sliva, IBM

 

An embedded DRAM macro fabricated in 65nm CMOS achieves 1.0GHz multi-banked operation at 1.0V yielding 584 Gbits/sec.  The array utilizes a 0.11um2 cell with 20fF deep trench capacitor and 2.2nm gate oxide transfer gate.  Concurrent refresh allows for high availability via a second bank address.  At-speed test and repair is accomplished with a new hierarchical BIST architecture.  Measured random cycle time exceeds 333MHz at 1.0V with functional operation from 750mV to 1.5V and densities up to 36.5Mbits

 

24.3

11:15 am

Electrically Programmable Fuse (eFUSE): From Memory Redundancy to Autonomic Chips (INVITED PAPER),  N. Robson, J. Safran, C. Kothandaraman, A. Cestero, X. Chen, R. Rajeevakumar, A. Leslie, D. Moy, T. Kirihata and S. Iyer, IBM

 

Electrical fuse (eFUSE) has become a popular choice to enable memory redundancy, chip identification and authentication, analog device trimming, and other applications. We will review the evolution and applications of electrical fuse solutions for 180nm to 45nm technologies at IBM, and provide some  insight into future uses in 32nm technology and beyond with the eFUSE as a building block for the autonomic chip of the future.