1. Comprehensive Interconnect Scaling
Circuit-system-interconnect cross-layer design and optimization for emerging scaled interconnects such as through-silicon-via (TSV), silicon carrier [2], and on-chip network interconnects [1,3-5].
Direction of future interconnect scaling
2. Concurrent Device-Circuit-System Design Automation Methodology
Design automation methodology for device-circuit-system co-optimization [5] and synthesis by combining analog circuit optimization [3, 4] from device level with high-level system design automation method.
CAD for system+circuit+device joint development
3. Post-CMOS Electronics
System-circuit-device cross-layer study and evaluation for the emerging post-CMOS elements such as nanophotonics, MEMS switch, carbon nano tube, grapheme, and nanowire.
cross-layer study for post CMOS technology
4. Emerging Diversified System Application
Invention of new electronic circuits and systems for health care equipments, pervasive computing systems, mobile computation units, lab-on-a-chip equipments, and scientific nano-/bio-measurement tools.
Emerging applications by combining different fields
Selected Past Research
1. Circuit Design
[1] Byungsub Kim and Vladimir Stojanović, “An Energy-Efficient Equalized Transceiver for RC-dominant Channels,” IEEE Journal of Solid-State Circuits, vol. 45, issue 6, pp. 1186-1197, June 2010.
- Improved data rate density and energy efficiency using equalizing transceiver circutis for a 10mm On-chip Interconnect.
- operated 4-6Gb/s/ch and 2-3Gb/s/um over a >40dB* channel loss.
- >40dB and low power operation is very rare.
- energy efficiency was about 0.4-0.6pJ/b
- Charge-injection equalization transmitter architecture is introduced for the first time
- For on-chip interconnect appplication, energy efficiency is imporved by 2x, 2x, and 4x compared to current switch, voltage mode, and current mode driver, respectively.
- For 50Ohm off-chiop interconnect application, energy efficiency is improved by 2x, 3x, and 8x compared to current switch, voltage mode, and current mode driver, respectively.
- Transmitter's resolution requirement is improved by 10x. This relaxation of resolution requirement enables >40dB-operation at cheap power and hardware cost.
- Relaxation of resolution requirement is analyzed by eye-sensitivity analysis and demonstrated in experiment.
- Charge-injection scheme alows cheap static pre-distortion
- Static pre-distortion technique is introduced for the first time.
- Static pre-distortion utilizes non-linear circuits as a linear feed forward equalizing (FFE) transmitter at cheap cost.
- Current source circuits are replaced with 5x smaller (in width) inverters for FFE operation.
- Eye is improved by pre-distortion technique
- With conventional transmitter circuits
- eye closes in theoretical calculation without pre-distortion
- With charge-injection transmitter circuits
- eye is 30% improved by pre-distortion in simulation
- Trans-impedance amplifing receiver is suggested in this application to improve signal amplitude by 3x at 50% smaller static power cost for the same bandwidth.
- 2009 ADI Outstanding Student Designer Award from MIT
- 2009 ISSCC highlight
[2] Byungsub Kim, Young Liu, Timothy O. Dickson, John F. Bulzacchelli, and Daniel J. Friedman, “A 10-Gb/s Compact Low-Power Serial I/O with DFE-IIR Equalization in 64nm CMOS,” IEEE Journal of Solid-State Circuits, vol. 44, issue 12, pp. 3526-3538, December 2009 [Invited].
- An equalizing transceiver is developed, designed and tested for a low power and high speed data communication through emerging on-package interconnect (system in package application) called silicon carrier.
- A decision feedback equalization (DFE) with infinite impulse reponse (IIR) filter, called DFE-IIR, is designed to reduce hardware (DFE-tap count) and power cost.
- 1st DFE-tap circuit's critical delay is improved by direct current comparator to allow a low power direction feedback at 10Gb/s
- Receiver sensitivity and timing margin for the critical path is improved by doulbe regenerating latches. Double regenerating latches provided CMOS-to-CMOL voltage level conversion, improving tail current's accuracy.
- Analysis and suggestion of termination policy for the the emerging silicon carrier channel.
- A major award will be given to this paper in 2011.
- ISSCC paper received Beatrice Winner Award for the Editorial Excellence in 2009.
- ISSCC 2009 highlight
- Major contents in this paper are applied for US and International patents.
[3] Byungsub Kim and Vladimir Stojanović, “Equalized Interconnect for On-Chip Network: Modeling and Optimization Framework,” in Proc. IEEE/ACM Int. Conf. on Computer-Aided Design, November. 2007, pp. 552-559.
- A modeling and framework methology is suggested for network-on-a-chip (NoC) interconnects using a low power equalization technique.
- Cross-layer design automation direction including mixed-signal circuit is suggested.
- Modeling, analysis, approximation, and algorithms for fast and easy estimation for equalized interconnects.
- 30,000 times faster estimation compared to SPICE simulation for link metric computation.
- equalized interconnects are better than repeated interconnects in comparison.
- equalized interconnect toolbox is developed and being prepared for source code open.
[4] Byungsub Kim and Vladimir Stojanović, “Characterization of Equalized and Repeated Interconnects for NoC Applications,” IEEE Design and Test of Computers, vol. 25, issue 5, pp. 430-439, October 2008 [Invited].
- NoC interconnect metrics are compared when optimized using repeater insertion and equalization techniques in aggressively scaled 32nm technoloy.
- A simple approximation method is suggested to evaluate delay-power-energy 3-dimensional trade-off cruve or repeated interconnects.
- NoC metrics are compared when equalized interconnects and repeated interconnects are used by using event-driven NoC simulator and equalized interconnect toolbox as a system-circuit cross-layer design approach.
- Various combination of NoC topologies and interconnect types (equalized and repeated interconnects) are evaluated and compared.
- An example of the possible system-circuit cross-alyer design automation.