Search

For high-volume production of 3D-stacked chips with through-silicon-vias (TSVs), wafer-scale bonding offers lower production cost compared with bump bond technology and is promising for interconnect pitches smaller than 5 µ using available tooling. Prior work has presented wafer-scale integration with tungsten TSV for low-power applications. This paper reports the first use of low-temperature oxide bonding and copper TSV to stack high performance cache cores manufactured in 45 nm Silicon On Insulator-Complementary Metal Oxide Semiconductor (SOI-CMOS) embedded DRAM (EDRAM) having 12 to 13 copper wiring levels per strata and upto 11000 TSVs at 13 µm pitch for power and signal delivery. The wafers are thinned to 13 µm using grind polish and etch. TSVs are defined post bonding and thinning using conventional alignment techniques. Up to four additional metal levels are formed post bonding and TSV definition. A key feature of this process is its compatibility with the existing high performance POWER7™ EDRAM core requiring neither modification of the existing CMOS fabrication process nor re-design since the TSV RC characteristic is similar to typical 100–200 µm length wiring load enabling 3D macro-to-macro signaling without additional buffering Hardware measurements show no significant impact on device drive and off-current. Functional test at wafer level confirms 2.1 GHz 3D stacked EDRAM operation.

We describe a single voltage supply, 1 MB cache subsystem prototype that integrates 2 GHz embedded DRAM (eDRAM) macros with on-chip word-line voltage supply generation , a 4 Kb one-time-programmable read-only memory (OTPROM) for redundancy and repair control, on-chip OTPROM programming voltage generation, clock generation and distribution, array built-in self-test circuitry (ABIST), user logic and pervasive logic. The eDRAM employs a programmable pipeline, achieving 1.8 ns latency, and features concurrent refresh capability.

As microprocessors enter the highly multi-core/multi-threaded era, higher density, lower latency embedded memory will be required to meet cache design needs. This paper describes a 500 MHz random cycle silicon on insulator (SOI) embedded DRAM macro which features a three-transistor micro sense amplifier, realizing significant performance gains over traditional array design methods. To address the realities of process integration, we describe the features and issues associated with integrating this DRAM into SOI technology, including deep trench processing and floating body effects. After a brief description of the macro architecture, details are provided on the three-transistor micro sense amplifier scheme, which is key to achieving a high transfer ratio with minimal area overhead. The paper concludes with hardware results and a summary.

The Blue Gene®/L chip is a technological tour de force that embodies the system-on-a-chip concept in its entirety. This paper outlines the salient features of this 130-nm complementary metal oxide semiconductor (CMOS) technology, including the IBM unique embedded dynamic random access memory (DRAM) technology. Crucial to the execution of Blue Gene/L is the simultaneous instantiation of multiple PowerPC® cores, high-performance static random access memory (SRAM), DRAM, and several other logic design blocks on a single-platform technology. The IBM embedded DRAM platform allows this seamless integration without compromising performance, reliability, or yield. We discuss the process architecture, the key parameters of the logic components used in the processor cores and other logic design blocks, the SRAM features used in the L2 cache, and the embedded DRAM that forms the L3 cache. We also discuss the evolution of embedded DRAM technology into a higher-performance space in the 90-nm and 65-nm nodes and the potential for dynamic memory to improve overall memory subsystem performance.

This work describes a 500-MHz compiled eDRAM macro offered in a 90-nm logic-based process. The macro architecture is optimized for high bandwidth while enabling compilation in bank and data-word dimensions. A direct write scheme simultaneously improves random bank cycle time and row access time without signal loss. The benefits of ground sensing, reference cells, and bitline twisting was reviewed. A variable stage pipeline extends the macro bandwidth while offering flexibility in clock frequencies. The redundancy system is modified to support direct write and piping. Finally, BIST was enhanced to utilize electrically blown fuses, enabling one-touch test and repair. Hardware results was presented.

This paper describes a 144-Mb DRAM that operates at a random cycle of 5.6 ns and is capable of producing data rates of 1.4 Gb/s/pin. The 121-mm/sup 2/ die is fabricated in a 0.13-/spl mu/m logic-based process with embedded DRAM. The cycle time is achieved using an early-write sensing technique that eliminates most of the timing overhead associated with the write cycle. Dynamic-precharge decoding in the subarray decode path is implemented to improve the access time. An improved data-formatting circuit is used to arrange the exit order of the eight-word burst. These circuit techniques produce latencies of 5.0 ns. The DRAM uses a DDR3-SRAM interface and is function and package compatible with industry-standard DDR3 SRAMs. Highlights of the DDR3 interface include the use of active termination circuitry on all inputs. The active termination improves the data-eye window and improves data capturing with minimum data setup and hold.

A built-in self-test engine and test methodology have been developed for testing a family of high-bandwidth, high-density DRAM macros. The DRAM macros range in size from 256/spl times/16/spl times/128 to 2 K/spl times/16/spl times/256 (Word/spl times/Bit/spl times/Data) and are targeted for embedded applications in application-specific integrated circuit designs. The processor-based test engine, with two separate instruction storage memories, combines with flexible address, data, and clock generators to provide DRAM high-performance ac testing using a minimum of dedicated test pins. Test results are compressed through on-macro, two-dimensional, redundancy allocation logic to provide direct programming information for the fuser via a serial scan port. The design is intended for reuse on future DRAM-generation subarrays and can be adapted to any number of address or data-pin configurations.

For high-volume production of 3D-stacked chips with through-silicon-via (TSVs), wafer-scale bonding offers lower production cost compared with bump bond technology [1][2][3] and is promising for interconnect pitch

The Isolated Preset Architecture (IPA) improves retention characteristics by implementing a weak read ‘1’ Isolation scheme, allowing a lower stored ‘1’ level to be sensed. The architecture also reduces sub-array area by 15% and bit-line activation power by 2× compared to previous design, without impacting performance. The architecture was implemented in IBM's 32nm High-K/Metal SOI embedded DRAM technology. Hardware results confirm 1.8ns random cycle and 2× improved retention characteristic with optimized Analog reference tuning.

In this paper, we present a systematic performance study and modeling of on-chip deep trench (DT) decoupling capacitors for high-performance SOI microprocessors. Based on system-level simulations, it is shown that the DT decoupling capacitors (decap) offer significant area advantage over the other two conventional decoupling capacitors - Metal-oxide-semiconductor (MOS) and Metal-Insulator-Metal (MIM). The fabrication process flow of DT decap is borrowed from regular eDRAM process and adds no additional process cost to processors that utilize large eDRAM cache [1]. We demonstrate that, with new process innovations such as introduction of High-k/metal gate and new plate doping methodology, there is significant reduction in equivalent series resistance (ESR) of the trench resulting in ∼3.5X improvement in half capacitance frequency for 32nm node. Further, with 22nm technology, improved ESR, DT Decaps performance is significantly enhanced, hence showing that DT-decaps can be reliably used for technology beyond 32nm.

This paper reviews the technology and design considerations for the implementation of 3 Dimensional integration of memory in a high performance logic environment

A technique for in-situ measurement of process variation in deep trench capacitance, bitline capacitance, and device threshold voltage in embedded DRAM arrays is presented. The technique is used to directly measure the parameter statistics in two product representative 45-nm SOI eDRAM arrays.

Logic-based embedded DRAM has matured into a wide range of ASIC applications, SRAM replacements [1] and off-chip caches for microprocessors [2]. While embedded DRAM has been leveraged in supercomputers such as IBM's BlueGene/L [3], it's use has been limited to moderate performance bulk logic technologies. Although prototypes have been demonstrated [4], DRAM has yet to be embedded on a high performance microprocessor. This paper discloses an SOI DRAM macro implemented on-chip with the IBM POWER7™ high performance microprocessor [5], and introduces enhancements to the micro sense amp (µSA) architecture [6]. This high performance DRAM macro is used to construct a large 32MB L3 cache on-chip, eliminating delay, area and power from the off-chip interface, simultaneously improving system performance, reducing cost, power and soft error vulnerability. Figure 19.1.1a shows an SEM of the 45nm SOI DRAM Device and Deep Trench (DT) capacitor [7]. DT offers 25x more capacitance than planar structures and was also utilized to reduce on-chip voltage island supply noise.

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 2 GHz AC array access and can endure excessive DC load.

We present a 1 MB cache subsystem that integrates 2 GHz embedded DRAM macros, charge pump circuits, a 4 Kb one-time-programmable ROM, clock multipliers, and built-in self test circuitry, having a 36.5 GB/s peak system data-rate. The eDRAM employs a programmable pipeline, achieving a 1.8 ns latency.

This tutorial will provide a bottom-up view of the changes in semiconductor memory design as we move into the nanometer regime. We begin by discussing the breakdown of scaling and the power problem. As innovation replaces classical scaling we investigate the use of stress engineering to improve device level performance. Technology challenges in lithography and interconnects are addressed. The consequences of innovation and scaling on RF/Analog characteristics must also be considered. The scaling of memory presents yet another challenge. We proceed to discuss the modeling of these effects for the circuit designer including discussion of the many new and traditional sources of variation. We describe how these are characterized how they can be controlled by layout rules and how the remaining variation can be describe in the model to enable Statistical Timing and other advanced circuit techniques. At the circuit level we consider in detail embedded DRAM and SRAM design for both bulk and SOI. We discuss the benefits and challenges of advanced technologies including methods for creating robust designs in the presence of manufacturing variation. We also discuss the design innovations required to utilize advanced technologies for overcoming the "memory wall", "power wall" and "ILP wall".

A high-speed 16-Mb DRAM chip with on-chip error-correcting code (ECC), which supports either 11/11 or 12/0 RAS/CAS addressing and operates on a 3.3- or 5-V power supply, is described. It can be packaged as a 2-Mb*8, 4-Mb*4, 8-Mb*2, or 16-Mb*1 DRAM, And is capable of operating in fast page mode, static column mode, or toggle mode. Speed and flexibility are achieved by a pipeline layout and on-chip SRAMs that buffer entire ECC words. The use of redundant word and bit lines in conjunction with the ECC produces a synergistic fault-tolerance effect. >

A prototype SOI embedded DRAM macro is developed for high-performance microprocessors and introduces performance-enhancing 3T micro sense amplifier architecture (muSA). The macro was characterized via a test chip fabricated in a 65nm SOI deep-trench DRAM process. Measurements confirm 1.5ns random access time with a 1V supply at 85deg and low voltage operation with a 600mV supply.

A 0.12 /spl mu/m growable eDRAM macro has GND sense, bit-line twisting, direct reference cell write, a flexible multi-banking protocol, and column redundancy to support multi-banking. The protocol supports simultaneous activate, read/write and pre-charge to three different banks. Hardware measurements verify 300 MHz operation, 6.6 ns tacc, and 10 ns trc.

A 16-Mb DRAM chip fabricated in a 0.5- mu m CMOS process using silicided polysilicon, double metal, and trench storage is described. It incorporates an architecture that supports either 11/11 or 12/10 RAS/CAS (row-address strobe/column-address-strobe) addressing. It is segmented to utilize bit redundancy of 2 lines/137-b lines/half quadrant, and separate word-redundant array of 24 lines/quadrant, providing any-for-any-word-line replacement within a quadrant. It has single-error-correct/double-error-detect (SEC/DED) error checking and correcting (ECC) Hamming odd-weight code/quadrant and either 5-V or 3-V operation. The design provides a RAS access of 50 ns with 16-ns fast-page access, 18-ns static column or toggle of 10 ns at 2.9 V and 85 degrees C. DRAM features are summarized. >