Data movement between the processing units and the memory in traditional von Neumann architecture is creating the “memory wall” problem. To bridge the gap, two approaches, the memory-rich processor (more on-chip memory) and the compute-capable memory (processing-in-memory) have been studied. However, the first one has strong computing capability but limited memory capacity/bandwidth, whereas the second one is the exact the opposite.To address the challenge, we propose DRISA, a DRAM-based Reconfigurable In-Situ Accelerator architecture, to provide both powerful computing capability and large memory capacity/bandwidth. DRISA is primarily composed of DRAM memory arrays, in which every memory bitline can perform bitwise Boolean logic operations (such as NOR). DRISA can be reconfigured to compute various functions with the combination of the functionally complete Boolean logic operations and the proposed hierarchical internal data movement designs. We further optimize DRISA to achieve high performance by simultaneously activating multiple rows and subarrays to provide massive parallelism, unblocking the internal data movement bottlenecks, and optimizing activation latency and energy. We explore four design options and present a comprehensive case study to demonstrate significant acceleration of convolutional neural networks. The experimental results show that DRISA can achieve 8.8× speedup and 1.2× better energy efficiency compared with ASICs, and 7.7× speedup and 15× better energy efficiency over GPUs with integer operations.CCS CONCEPTS• Hardware → Dynamic memory; • Computer systems organization → reconfigurable computing; Neural networks;

DRISA: a DRAM-based Reconfigurable In-Situ Accelerator

Building programmable resistive devices in contact holes at the crossover of a plurality of conductor lines in more than two vertical layers is disclosed. There are plurality of first conductor lines and another plurality of second conductor lines that can be substantially perpendicular to each other, though in two different vertical layers. A diode and/or a programmable resistive element can be fabricated in the contact hole between the first and second conductor lines. The programmable resistive element can be coupled to another programmable resistive device or shared between two programmable devices whose diodes conducting currents in opposite directions and/or coupled to a common conductor line. The programmable resistive memory can be configured to be programmable by applying voltages to conduct current flowing through the programmable resistive element to change its resistance for a different logic state.

Programmable Resistive Device and Memory Using Diode as Selector

Decreasing read cell current (ICELL) has become a major trend in nonvolatile memory (NVM). However, a reduced ICELL leaves the operation of the sense amplifier (SAs) vulnerable to bitline (BL) level offset and SA input offset. Thus, small- ICELL NVMs suffer from slow read speed or low read yield. In this study, we propose a new current-sampling-based SA (CSB-SA) to suppress the offset due to device mismatch, while maintaining tolerance for insufficient precharge time. These features enable CSB-SA to achieve a read speed 6.3 ×-8.1× faster than previous SAs, for sensing 100 nA ICELLs on a 2 K-cell bitline. We fabricated a CMOS-logic-compatible, 90 nm, 512 Kb OTP macro, using the CSB-SA. This OTP macro achieves a random access time of 26 ns for reading sub-200 nA ICELL. Measurements confirm that this 90 nm CSB-SA is also capable of sub-100 nA sensing.

An Offset-Tolerant Fast-Random-Read Current-Sampling-Based Sense Amplifier for Small-Cell-Current Nonvolatile Memory

Junction diodes or MOS devices fabricated in standard FinFET technologies can be used as program selectors or One-Time Programmable (OTP) element in a programmable resistive device, such as interconnect fuse, contact/via fuse, anti-fuse, or emerging nonvolatile memory such as MRAM, PCRAM, CBRAM, or RRAM. The MOS or diode can be built on at least one fin structure or at least one active region that has at least one first active region and a second active region. The first and the second active regions can be isolated by a dummy MOS gate or silicide block layer (SBL) to construct a diode.

Circuit and system of using FinFET for building programmable resistive devices

A method and system for multiple-bit programmable resistive cells having a multiple-bit programmable resistive element and using diode as program selector are disclosed. The first and second terminals of the diode having a first and second types of dopants can be fabricated from source/drain of MOS in a well for MOS devices or fabricated on the same polysilicon structure. If a multiple-bit programmable resistive cell has 2 n (n>1) distinct resistance levels to store n-bit data, at least 2 n −1 reference resistance levels can be designated to differential resistances between two adjacent states. Programming multiple-bit programmable resistive elements can start by applying a program pulse with initial program voltage (or current) and duration. A read verification cycle can follow to determine if the desirable resistance level is reached. If the desired resistance level has not been reached, additional program pulses can be applied.

Multiple-Bit Programmable Resistive Memory Using Diode as Program Selector

This paper describes a second-generation one-time programmable read-only memory (OTPROM) that provides these features through a balanced bitline, resistor pull-up, differential sense amp with a programmable reference.

A Commercial Field-Programmable Dense eFUSE Array Memory with 99.999% Sense Yield for 45nm SOI CMOS

A 1.1 Mb embedded DRAM macro (eDRAM), for next-generation IBM SOI processors, employs 14 nm FinFET logic technology with $\hbox{0.0174}~\mu\hbox{m}^{2}$ deep-trench capacitor cell. A Gated-feedback sense amplifier enables a high voltage gain of a power-gated inverter at mid-level input voltage, while supporting 66 cells per local bit-line. A dynamic-and-gate-thin-oxide word-line driver that tracks standard logic process variation improves the eDRAM array performance with reduced area. The 1.1 $~$ Mb macro composed of 8 $\times$ 2 72 Kb subarrays is organized with a center interface block architecture, allowing 1 ns access latency and 1 ns bank interleaving operation using two banks, each having 2 ns random access cycle. 5 GHz operation has been demonstrated in a system prototype, which includes 6 instances of 1.1 Mb eDRAM macros, integrated with an array-built-in-self-test engine, phase-locked loop (PLL), and word-line high and word-line low voltage generators. The advantage of the 14 nm FinFET array over the 22 nm array was confirmed using direct tester control of the 1.1 Mb eDRAM macros integrated in 16 Mb inline monitor.

A 14 nm 1.1 Mb Embedded DRAM Macro With 1 ns Access

IBM introduced trench capacitor eDRAM into its high performance microprocessors beginning with 45nm and Power 7 [1] to provide a higher density cache without chip crossings. Whereas the 45 and 32nm designs employ a micro sense amplifier [2] and three-level bitline hierarchy, the design implemented for 22nm utilizes a higher gain sense amplifier and two-level bitline architecture that together provide significant reductions in area, latency, and power. This 22nm design style has been migrated into a 14nm FinFET [3] learning vehicle, complete with an ABIST engine, wordline charge pumps (VPP and VWL), and padcage interface circuitry.

17.4 A 14nm 1.1Mb embedded DRAM macro with 1ns access

Scaling trends in the alpha-particle and neutron induced SRAM SER shows an increase in the per-bit SER and percent multi-cell upsets at the 5-nm FinFET process compared to the 7-nm process. Neutron SER tests across process corners show that the faster process corner SER is up to $2\times$ higher than the slower process corner SER in 7-nm and 5-nm FinFETs. The process corner dependence of SER is attributed to differences in propagation delay and single-event transient pulse-widths.

Chris Paone

Papers

A Commercial Field-Programmable Dense eFUSE Array Memory with 99.999% Sense Yield for 45nm SOI CMOS

A 14 nm 1.1 Mb Embedded DRAM Macro With 1 ns Access

17.4 A 14nm 1.1Mb embedded DRAM macro with 1ns access

Scaling Trends and the Effect of Process Variations on the Soft Error Rate of advanced FinFET SRAMs