P. Thomas

5.9 Haswell: A family of IA 22nm processors

The 4th Generation Intel® Core™ processor, codenamed Haswell, is a family of products implemented on Intel 22nm Tri-gate process technology [1]. The primary goals for the Haswell program are platform integration and low power to enable smaller form factors. Haswell incorporates several building blocks, including: platform controller hubs (PCHs), memory, CPU, graphics and media processing engines, thus creating a portfolio of product segments from fan-less Ultrabooks™ to high-performance desktop, as shown in Fig. 5.9.1. It also integrates a number of new technologies: a fully integrated voltage regulator (VR) consolidating 5 platform VRs down to 1, on-die eDRAM cache for improved graphics performance, lower-power states, optimized IO interfaces, an Intel AVX2 instruction set that supports floating-point multiply-add (FMA), and 256b SIMD integer achieving 2× the number of floating-point and integer operations over its predecessor. The 22nm process is optimized for Haswell and includes 11 metal layers (2 additional metal layers vs. Ivy Bridge [2]), high-density metal-insulator-metal (MIM) capacitors, and is tuned for different leakage/speed targets based on the market segment. For example, in some low-power products, the process is optimized to reduce leakage by 75% at Vmin, while paying only 12% intrinsic device degradation at the high-voltage corner.

A CMOS 10-gb/s SONET transceiver

This paper presents a single-chip SONET OC-192 transceiver (transmitter and receiver) fabricated in a 90-nm mixed-signal CMOS process. The transmitter consists of a 10-GHz clock multiplier unit (CMU), 16:1 multiplexer, and 10-Gb/s output buffer. The receiver consists of a 10-Gb/s limiting input amplifier, clock and data recovery circuit (CDR), 1:16 demultiplexer, and drivers for low-voltage differential signal (LVDS) outputs. Both transmit and receive phase-locked loops employ a 10-GHz on-chip LC voltage-controlled oscillator (VCO). This transceiver exceeds all SONET OC-192 specifications with ample margin. Jitter generation at 10.66-Gb/s data rate is 18 mUI/sub pp/ (unit interval, peak-to-peak) and jitter tolerance is 0.6 UI/sub pp/ at 4-MHz jitter frequency. This transceiver requires 1.2V for the core logic and 1.8 V for input/output LVDS buffers. Multiple power supply domains are implemented here to mitigate crosstalk between receiver and transmitter. The overall power dissipation of this chip is 1.65 W.

Haswell: A Family of IA 22 nm Processors

We describe the 4th Generation Intel® Core™ processor family (codenamed “Haswell”) implemented on Intel® 22 nm technology and intended to support form factors from desktops to fan-less Ultrabooks™. Performance enhancements include a 102 GB/sec L4 eDRAM cache, hardware support for transactional synchronization, and new FMA instructions that double FP operations per clock. Power improvements include Fully-Integrated Voltage Regulators ( ~ 50% battery life extension), new low-power states (95% standby power savings), optimized MCP I/O system (1.0-1.22 pJ/b), and improved DDR I/O circuits (40% active and 100x idle power savings). Other improvements include full-platform optimization via integrated display I/O interfaces.

Optical I/O technology for digital VLSI

We describe the development of a high-speed, 12-channel (8-data, 2-clock and 2-alignment channels), parallel optical link with a unique packaging concept. The package is used to demonstrate the viability of chip-to-chip optical I/O in very large scale integration (VLSI) circuits. However, for implementation of optical systems in high performance computing applications, the cost of components and packaging has to come down significantly from the traditional optical communication distances. In the current work we attempted to realize such a system by using power efficient optical and electronic components together with a potentially low cost packaging solution compatible with the electronics industry. Vertical Cavity Surface Emitting Lasers (VCSEL), positive-intrinsic-negative (PIN) photodetectors, polymer waveguide arrays as well as CMOS transceiver chip were heterogeneously integrated on a standard microprocessor flip-chip pin grid array (FCPGA) substrate. The CMOS transceiver chip from 0.18μm processing technology contains VCSEL drivers, transimpedance and limiting amplifiers and on-chip self-testing circuits. A self-test circuit in such high-speed systems will be highly beneficial to reduce the testing cost in real products. For fully assembled packages we measured a 3 Gb/s optical eye for the transmitter (24Gb/s aggregate data rate) and a transmission over the complete link was achieved at 1 Gb/s (8Gb/s aggregate data rate).

Optical hybrid package with an 8-channel 18GT/s CMOS transceiver for chip-to-chip optical interconnect

We describe the design and development of a high-speed 8-channel hybrid integrated optical transceiver package with Clock and Data Recovery (CDR) circuits. The package concept has been developed to be compatible with microprocessor package technology and at the same time allow the integration of low cost, high-performance optical components. A 90nm CMOS optical transceiver chip, 850nm 10Gb/s GaAs based vertical cavity surface emitting laser (VCSEL) array and PIN photodiode array are flip-chip mounted on a standard microprocessor Land Grid Array (LGA) package substrate. The CMOS drivers and receivers on the transceiver chip and the optical components (VCSEL and Photodiode arrays) are electrically coupled using a short transmission line routed on the top surface of the package. VCSEL and photodiode arrays are optically coupled to on-package integrated polymer waveguide arrays with metallized 45° mirrors. The waveguides, which are terminated with multi-terminal (MT) fiber optic connectors, couple out/in high-speed optical signals to/from the chip. The CMOS transceiver chip fully integrates all analog optical circuits such as VCSEL drivers, transimpedance amplifiers and clock and data recovery (CDR) retiming circuit with a low jitter LC-PLL. Digital circuits for pseudorandom bit-pattern sequence generators (PRBS) and bit-error rate test (BERT) are fully integrated. 20Gb/s electrical and 18Gb/s optical eye diagrams for the transmitter were measured out of the package. A fully packaged transmitter and receiver including clock data recovery at 10Gb/s have also been measured.

Four-way processor 800 MT/s front side bus with ground referenced voltage source I/O

A 40 cm multi-drop bus shared by 5 test chips to emulate 4 processors and a chipset runs error free at 800 MT/s with 130 mV margin using Ground Referenced Voltage Source (GRVS) I/O scheme. For comparison, when the same test chip is programmed to use Gunning Transceiver Logic (GTL), the bus speed is 500 MT/s for the same 130 mV margin under identical conditions.