Search

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.

A 128-kb magnetic random access memory (MRAM) test chip has been fabricated utilizing, for the first time, a 0.18-/spl mu/m V/sub DD/=1.8 V logic process technology with Cu metallization. The presented design uses a 1.4-/spl mu/m/sup 2/ one-transistor/one-magnetic tunnel junction (1T1MTJ) cell and features a symmetrical high-speed sensing architecture using complementary reference cells and configurable load devices. Extrapolations from test chip measurements and circuit assessments predict a 5-ns random array read access time and random write operations with

With the promise of nonvolatility, practically infinite write endurance, and short read and write times, magnetic tunnel junction magnetic random access memory could become a future mainstream memory technology.

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 study the yield of a 65nm SOI eDRAM design. The impact of random dopant fluctuations in the cell and micro sense amp is studied under different systematic corner and device type considerations. Trench capacitor variation effects and yield timing windows are evaluated. By analyzing the circuit hierarchy, we are able to understand yield trends and hence suitable design guidelines. For the first time a fast montecarlo statistical analysis approach is employed for eDRAM analysis. For this, we rely on a well-developed SRAM statistical yield analysis methodology. It is shown that a multiplicity of devices and design considerations can be critical based on the memory operating corners and conditions. It is also shown that high vt sense amp devices, unlike low Vt devices, enable sufficient design yield windows.

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.

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.

The disciplines of process development, circuit design, and microarchitecture too often remain isolated. This paper on memories explores the benefit of their unification by taking a retrospective look at a L1 cache memory design, a current view of embedded DRAM, and a speculative peek at emerging memories (e.g. MTJ MRAM). In the midst, a novel refresh operation is proposed for DRAM that relies on read and write activity ongoing within a cache to refresh the DRAM.

Magnetic random access memory (MRAM) offers an alternative approach to fast low-power non-volatile VLSI memory. MRAM has been pursued for more than 10 years as a robust non-volatile memory for space applications. The magnetic tunnel junction (MTJ) MRAM is dramatically different and achieves four orders of magnitude better bandwidth to sense power ratio by utilizing a high resistance and high magneto-resistance (MR) MTJ and including a FET switch in each cell. Non-volatile storage and 10 ns performance are demonstrated in 1 kb arrays. Read and write on-chip power at 2.5 V and 100 MHz are 5 mW and 40 mW respectively.

Chip performance, power, noise, and clock synchronization are becoming formidable challenges as microprocessor performance moves into the GHz regime and beyond. Interlocked pipelined CMOS (IPCMOS), an asynchronous clocking technique, helps address these challenges. This paper shows how a typical block (e.g., Block D) is interlocked with all the blocks with which it interacts. In the forward direction, dedicated Valid signals emulate the worst-case path through each driving block and thus determine when data can be latched within the typical block. In the reverse direction, Acknowledge signals indicate that data has been received by the subsequent blocks and that new data may be processed within the typical block. In this interlocked approach local clocks are generated only when there is an operation to perform.