Ivan Vo

A million spiking-neuron integrated circuit with a scalable communication network and interface

Modeling computer chips on real brains Computers are nowhere near as versatile as our own brains. Merolla et al. applied our present knowledge of the structure and function of the brain to design a new computer chip that uses the same wiring rules and architecture. The flexible, scalable chip operated efficiently in real time, while using very little power. Science, this issue p. 668 A large-scale computer chip mimics many features of a real brain. Inspired by the brain’s structure, we have developed an efficient, scalable, and flexible non–von Neumann architecture that leverages contemporary silicon technology. To demonstrate, we built a 5.4-billion-transistor chip with 4096 neurosynaptic cores interconnected via an intrachip network that integrates 1 million programmable spiking neurons and 256 million configurable synapses. Chips can be tiled in two dimensions via an interchip communication interface, seamlessly scaling the architecture to a cortexlike sheet of arbitrary size. The architecture is well suited to many applications that use complex neural networks in real time, for example, multiobject detection and classification. With 400-pixel-by-240-pixel video input at 30 frames per second, the chip consumes 63 milliwatts.

A 1.0-GHz single-issue 64-bit powerPC integer processor

This 64 b single-issue integer processor, comprised of about one million transistors, is fabricated in a 0.15 /spl mu/m effective channel length, six-metal-layer CMOS technology. Intended as a vehicle to explore circuit, clocking, microarchitecture, and methodology options for high-frequency processors, the processor prototype implements 60 fixed-point compare, logical, arithmetic, and rotate-merge-mask instructions of the PowerPC instruction-set architecture with single-cycle latency. The processor executes programs written in this instruction subset from cache with a 1 ns cycle. In addition, the prototype implements 36 PowerPC load/store instructions that execute as single-cycle operations (zero wait cycles) with 1.15 ns latency. Full data forwarding and full at speed scan testing are supported.

Design methodology for a 1.0 GHz microprocessor

This paper describes the design methodology used to build an experimental 1.0 GigaHertz PowerPC integer microprocessor at IBMs Austin Research Laboratory. The high frequency requirements dictated the chip composition to be almost entirely custom macros using dynamic circuit techniques. The methodology presented will cover design and verification tools as well as circuit constraints and microarchitecture philosophy. The microarchitecture, circuits and tools were defined by the high frequency requirements of the processor as well as the aggressive design schedule and size of the design team.

Real-time scalable cortical computing at 46 giga-synaptic OPS/watt with ~100× speedup in time-to-solution and ~100,000× reduction in energy-to-solution

Drawing on neuroscience, we have developed a parallel, event-driven kernel for neurosynaptic computation, that is efficient with respect to computation, memory, and communication. Building on the previously demonstrated highly optimized software expression of the kernel, here, we demonstrate True North, a co-designed silicon expression of the kernel. True North achieves five orders of magnitude reduction in energy to-solution and two orders of magnitude speedup in time-to solution, when running computer vision applications and complex recurrent neural network simulations. Breaking path with the von Neumann architecture, True North is a 4,096 core, 1 million neuron, and 256 million synapse brain-inspired neurosynaptic processor, that consumes 65mW of power running at real-time and delivers performance of 46 Giga-Synaptic OPS/Watt. We demonstrate seamless tiling of True North chips into arrays, forming a foundation for cortex-like scalability. True Norths unprecedented time-to-solution, energy-to-solution, size, scalability, and performance combined with the underlying flexibility of the kernel enable a broad range of cognitive applications.

An 8GHz floating-point multiply

The implementation of the mantissa portion of a floating-point multiply (54/spl times/54b) is described. The 0.124mm/sup 2/ multiplier is implemented using limited switch dynamic logic and operates at speeds up to 8GHz in a 90nm SOI technology. The multiplier dissipates between 150mW and 1.8W as it scales between 2GHz and 8GHz.

A semi-custom voltage-island technique and its application to high-speed serial links

Supply-voltage reduction is a known technique for reducing CMOS active power. We propose a semi-custom voltage-island approach based on internal regulation and selective custom design. This approach enables transparent embedding, since no additional external power supply is needed. We apply the approach to high-speed serial links, and we show that high performance is retained through targeted application of custom circuit and logic design. A chip is presented that evaluates the presented approach on a 3000-gate 3.2-Gbps multi-protocol serial-link receiver logic core. When reducing the supply from 1.2V to 0.95V, the chip demonstrates power savings of over 25%.

On-chip jitter and oscilloscope circuits using an asynchronous sample clock

We demonstrate digital circuits for measuring the jitter histograms of gigahertz clock and data signals. The circuits do not require calibration, and an asynchronous sampling technique alleviates the need for an on-chip sample clock generator with delay control. We combine measurements across swept reference voltages to create statistical clock signal and eye diagram waveforms at 6GHz and 5Gbit/s, respectively. The proposed technique produced RMS jitter measurements of 2.0ps on clock signals and 6.2ps on random data signals.

An on-chip dual supply charge pump system for 45nm PD SOI eDRAM

We present an on-chip word line (WL) dual supply system for server class embedded DRAM (eDRAM) applications. The design consists of switched capacitor charge pumps, voltage regulators, reference and clock circuits. Charge pump engines feature efficient charge transfer and energy conversion, boosting unregulated rails to 1.8x supply. At vdd=1 V, regulated high (1.5 to 1.7 V) and low (-0.3 to -0.6 V) levels ensure WL overdrive and cell turn-off, respectively, with rippling <plusmn35 mV and maintenance power <780 muW/2Mb-DRAM. The system supports >2 GHz AC array access and can endure excessive DC load.

A 1V 18GHz Clock Generator in a 65nm PD-SOI Technology

Two PLLs were designed using current-steering interpolating ring oscillators. The regular-V, PLL demonstrates a 3.2times lock range and a maximum frequency of 24.6GHz with 1.28psrms jitter at 1V. The high-V, PLL exhibits a 3.5times lock range at 6% lower frequency. The 0.18mm2 PLLs consume 16mW of power from 1V and are fabricated in a PD-SOI 65nm technology

A semi-custom voltage-island technique and its application to high-speed serial links [CMOS active power reduction]

Supply-voltage reduction is a known technique for reducing CMOS active power. We propose a semi-custom voltage-island approach based on internal regulation and selective custom design. This approach enables transparent embedding, since no additional external power supply is needed. We apply the approach to high-speed serial links, and we show that high performance is retained through targeted application of custom circuit and logic design. A chip is presented that evaluates the presented approach on a 3000 gate 3.2 Gbps multi-protocol serial-link receiver logic core. When reducing the supply from 1.2 V to 0.95 V, the chip demonstrates power savings of over 25%.