DIGITAL
- Daytime Paper Sessions
- Evening Panel Discussion
- Tutorial
- Trend Charts
1998 ISSCC - DIGITAL
Subcommittee Chair:
Ian Young, Intel Corporation, Hillsboro, OR.
HIGHLIGHTS
- Next-generation x86 microprocessor designs [FP15]
- Architecture, circuits, and process technology for 1GHz CMOS Microprocessors [FP15, SP25]
- Microprocessors for media-intensive embedded applications [SA18]
MOST-SIGNIFICANT RESULTS
- First demonstration of a 1GHz microprocessor [15.1]
- First microprocessor with copper interconnect [15.6]
- First dynamic multi-threaded microprocessor [15.3]
- Third-generation 450MHz P6 microprocessor [15.4]
- 667MHz commercial microprocessor [18.4]
- 800Mbit/s per pin interface to high-density memories [10.1, 10.4]
- 1.2GHz phase-locked loop for a 600MHz Alpha microprocessor [25.2]
- Low-skew GHz clock distribution schemes [25.3, 25.4, 25.5
- MPEG2 (ML@MP) decoding in software at 1.2W [18.2, 18.6]
APPLICATIONS
- High-Performance Personal Computing and Servers [FA10, FP15, SP25]
- Higher-Performance, Lower Power, Portable Computing [FP15, SA18]
- Portable and Embedded Multimedia [SA18]
- PDAs [18.3]
- Video-on-demand set-top boxes [18.6]
ECONOMIC AND SOCIAL IMPACT
- Increased use of digital audio, graphics, and video content in daily life
- Improved human interface - handwriting and speech recognition
- Business productivity
- Flexible work environment / telecommuting
PANEL
- Will Power Limit Microprocessor Performance? [TE4]
TUTORIAL
- High-Speed Clocking for Large Digital ICS [T6]
HOT TOPICS
---Why Copper Interconnect? [15.6]
- Background
- Deep sub-micron interconnect is becoming increasingly resistive due to very narrow wires with larger R
- Coupling capacitance to adjacent wires is increasing due to narrow spacing between wires with larger C
- RC wire delays do not scale well with shrinking dimensions
- Copper has 40% lower resistance than Aluminum leading to reduced RC wire delays
- Challenges
- Copper interconnect is difficult to manufacture. It requires barrier layers to stop copper diffusing into the silicon. It is difficult to pattern. These problems have now been overcome.
- Copper interconnect will pose a challenge for further reducing RC delay. Thinner barrier layers and low-k dielectrics will be the next challenges
---What is multi-threading? [15.3]
- Multi-threading is a technique for improving computer performance.
- There are several different ways to do this, either in software or in hardware, either course-grain or fine-grain.
- This year ISSCC has a paper describing the first microprocessor implementing a mainstream architecture (the PowerPC) with course-grain multithreading.
- Course-grain multithreading hides the delays caused by data fetches from memory by switching between 2 or more program threads.
- When required data is not found in the on-chip cache, the processor will start fetching instructions from a new thread.
- Otherwise-idle processor cycles can now be filled with useful computation, thus improving performance
Session:
FA10 Subcommittee: Digital, Memory, TD
High-Speed Chip-to-Chip Connections
Chair:
Jeff Yetter, Hewlett Packard Co., Fort Collins, CO.
Associate Chair:
Bruce Bateman, MicroUnity Systems Engineering, Sunnyvale, CA.
DRIVERS
- Higher bandwidth for increased throughput
- Lower power for reduced system cost
HIGHLIGHTS
- 800Mb/s per pin interface [10.1, 10.4]
- Data driven on both clock edges [10.1, 10.4]
- Comparison of competing standards:
- Data synchronized to master clock [10.1, 10.4, 10.5]
- Clock/strobe sourced with data [10.2, 10.3]
- 320Mb/s per pin interface [10.2]
- 600Mb/s per pin interface [10.3]
- 1Gb/s at 10mW per pin interface [10.5]
- Low-skew global clock-distribution scheme [10.5]
Session:
FP15 Subcommittee: DIGITAL
Microprocessors
Chair: Ted Williams AMD, Milpitas, CA.
Associate Chair: Vojin G. Oklobdzjia, Integration, Berkeley, CA.
DRIVERS
- General-purpose computing
- Low-power-portable computers
- Personal Computers
- Workstations
HIGHLIGHTS
- First demonstration of a 1GHz microprocessor [15.1]
- 600MHz out-of-order instruction-queue circuit [15.2]
- First multi-threaded PowerPC microprocessor [15.3]
- Third-Generation 450MHz Intel P6 microprocessor [15.4]
- Highly-integrated low-power 200MHz, 0.5W Strong Arm microprocessor [15.5]
- First microprocessor with copper interconnect [15.6]
- 0.25micron AMD K6 microprocessor with 100MHz socket 7 [15.7]
Session:
SA18 Subcommittee: SIGNAL PROCESSING/DIGITAL
MultiMedia Processors and Elements
Chair:
Steve Purcell, Chromatic Research, Sunnyvale, CA.
Associate Chair: Tomohisa Arai, NEC Corporation, Kanagawa, Japan
DRIVERS
- Low Power for Portable and Embedded Applications
- Flexible Software-based Processing
- High-Performance Arithmetic
- Multiple Forms of Parallelism for Multimedia
HIGHLIGHTS
- 1.4GFLOP processor with graphics vector-floating-point unit [18.1]
- MPEG2 (ML@MP) decoding in software at 1.2W [18.2,18.6]
- 110mW, 1.5V, 4-way-parallel DSP for portable multimedia [18.3]
- Fastest commercial microprocessor : 667MHz [18.4]
- 2.7ns multiplier array for a 400MHz FPU [18.5]
Session:
SP25 Subcommittee: DIGITAL
Clock Networks
Chair: Dennis Cox, IBM Corporation, Rochester, MN.
Associate Chair: Gian Gerosa, Motorola Inc., Austin, TX.
DRIVERS
- GHz clock distribution
- Phase-locked loops (PLL)
- Delay-locked loops (DLL)
- High frequency
- Low jitter
- Low skew
- Skew immunity
- Race-free operation
HIGHLIGHTS
- 1.2GHz phase-locked loop for a 600MHz Alpha microprocessor [25.1]
- Clock design for a 600MHz Alpha microprocessor [25.2]
- Low-skew GHz clock distribution schemes [25.3, 25.4, 25.5]
- Clock distribution network for a 450MHz P6 [25.3]
- Race-free impulse latch circuits [25.4]
Panel Session:
TE4 Subcommittee: Digital
Will Power Limit Microprocessor Performance?
OBJECTIVE
- To discuss the practical limits of microprocessor power consumption and to propose ways in which performance can continue its exponential growth
APPLICATIONS
CHALLENGES
- Thermal management and reliability
- Power-supply distribution and signal integrity
CONTROVERSIES
- Architectural impact on performance/power tradeoffs
- Clock gating and micro-architectural solutions
- Is Vdd scaling a solution?
- Are there process-technology solutions?
- Are packaging breakthroughs to be expected?
- To what extent can low-power circuit-design techniques be applied to high-performance microprocessor design?
Tutorial
: T6 Subcommittee: Digital
High-Speed Clocking for Large Digital ICs
John Maneatis
OVERVIEW
- Clock generation and distribution
- Clock-skew budgeting and management
- Trade-offs in latch and register design
- Setup and hold-time constraints for state elements
- Various logic-pipeline approaches
- Clock-gating issues
TUTORIAL SPEAKER BIOGRAPHY
John Maneatis received the BS in EECS from UC Berkeley, in 1988 and the MS and PhD in EE from Stanford, in 1989 and 1994. At Stanford, his research interests included high-performance circuit design for phase-locked loops, microprocessors, data conversion, and clock recovery. Since 1994 he has been a circuit designer at Silicon Graphics, Mountain View, CA, in microprocessor design, clocking, and phase-locked loops.
Go back to the SSCS page
Go back to the ISSCC page
If you have any comments for the ISSCC, please forward them to
Frank Hewlett
hewletfw@sandia.gov
Comments related to the maintenance of this web site should be sent to
sscs@eecg.toronto.edu .
http://www.isscc.org/isscc/1998/press/digital.htm
Last modified : Tuesday February 17, 1998 at 8:12am EST