1 Options for Radiation Tolerant High- Performance Memory Steven M. Guertin, Member, IEEE, Jean Yang-Scharlotta, Member, IEEE, Raphael Some • Identify impact of RHBD limitations, especially at low Abstract— Memory technologies were reviewed for radiation feature size. effects performance in order to determine the most cost-efficient • For each memory technology, identify cell-level radiation target for a possible memory investment targeted at creating a performance strengths and weaknesses. memory component of general applicability for space use. TID • Identify specific trends in memory controller strengths and SEL trends that had indicated improvement in intrinsic performance based on scaling cannot be trusted. SEE hardening and weaknesses (i.e. Flash memory charge pumps). of memory arrays is prohibitive in terms of cell-level power and • Identify any other strengths or weaknesses of note for area requirements. Radiation-hardened cells and custom memory technologies. controllers are recommended for future memory developments. • Ensure the memory technologies are single-event latchup (SEL) free. I. INTRODUCTION • Identify memory technologies that can provide DRAM- pace missions require memory for many different like performance and/or Flash-memory like performance, Sapplic ations, including computer main memory, buffers, while meeting single- and multiple-bit upset (SBU and MBU) program and data storage, boot ROMs (read only memory), and SEFI rate requirements (with SEFI being the more and other applications. In the specific cases of high density problematic). volatile memory (currently handled with dynamic random- The approach taken for this work was the following. We access memory [DRAM] devices), and high density non- took two generic space environments – one that was heavy-ion volatile memory (currently handled with Flash memory), it dominated, and one that was proton dominated. We identified was determined that total ionizing dose (TID) and single-event the available RHBD suppliers and evaluated the performance effects (SEE) performance were insufficient for high- of their RHBD libraries. We then identified memory reliability deep space systems. A comprehensive review of technologies and their radiation performance and trends. With available memory technologies, coupled with a review of their the last several generations of performance as a starting point, radiation performance, will help identify the memory we projected technology trends into the future for memory technologies with the most potential to benefit from a modest devices. effort to develop a next-generation radiation hardened memory This summary is organized as follows. Background is technology. presented in Section II. Environments considered for this work Radiation performance trends for technologies and devices are presented in Section III. A review of RHBD programs is provide key information about current and future expectations. presented in Section IV. Reviews of individual memory trends Potential high-performance radiation hardened memory is presented in Section V. An exploration of packaging impact devices include radiation hardened by design devices, on radiation response is presented in Section VI. And finally, commercial devices, and hybrids. In this paper, we review the findings are presented in Section VII. radiation hardened by design (RHBD) libraries as well as fabrication options, expected performance, and limitations. In II. BACKGROUND addition, we provide a technology review of a range of The following general assumptions are made in order to advanced memory technologies and their expected radiation focus this work: performance into the future. These analyses provide a basis 1) The general timeframe targeting a future improved from which it will be possible to identify if high-performance product is about 2-4 years. Significant findings that seem to deep space memory requirements can be met now, in the require more than 4 years are noted but not pursued here. Our future, or in the future with the addition of targeted focus in then on memory technology predictions in the next development to remedy specific shortcomings of predicted few years only. performance. 2) Device densities of 256 Mb or larger are required for deep This paper is a formal review of RHBD and commercial space missions. Smaller density devices (and developing cell technologies of potential use for development of a space technologies) are not fully explored. memory. The goals of this paper, are to: 3) Feature size must be 65 nm or smaller for the required • Use appropriate radiation environments for evaluation of density. radiation effects in memory devices and technologies. 4) The types of memories of interest are limited to the • Review available RHBD options and assess radiation following: static random-access ram (SRAM), dynamic RAM performance. (DRAM), NAND Flash, NOR Flash, magnetic RAM 2 (MRAM), spin-torque transfer (STT), ferro-electric RAM radiation hardening by design. That is, by just informing the (FeRAM), phase change RAM (PCRAM), and resistive RAM fabrication facility to construct a device differently, radiation (RRAM). performance can be improved (as opposed to modifying the 5) The types of radiation effects of the greatest concern are actual process the facility uses, which has been termed the following: TID, SBU, MBU, SEFI, SEL, Stuck Bits, and “radiation hardening by process”). There are a handful of Other types of permanent failures suppliers that can create new integrated circuits (ICs) or 6) We identified two use categories of the greatest concern modify their commercial designs, using RHBD methods. – high speed memory to act as computer main memory, and For this effort, it is important to be informed of RHBD- high density non-volatile memory to act as file system storage. supplied memories and how they fit requirements. RHBD capabilities and available densities (related to feature size) are Figure 1: Commercial trends and methods to improve hardness, per Amort [1]. There are some notable examples that deviate from the predicted SEE and TID trend lines. Also, this is changing in the sub-32 nm range as mixed-signal and process changes are making this only a rough guideline. also important. With this information, it is possible to establish the state of the art for memory constructed using RHBD III. ENVIRONMENTS methods, where those efforts are likely to lead in the future, For this paper, we focused on technology performance and whether a custom development using RHBD is feasible. under a subset of key mission environments. These are the In this review, we explore the rad hard by design programs following: low-earth orbit (LEO) at 51.6° inclination is the listed below. Each of the programs has been reviewed and the International Space Station orbit (ISS) [2]; geo-stationary orbit findings are discussed. (GEO); and interplanetary space. GEO orbit includes a portion 1) Boeing 90 nm – as a baseline of the upper Van-Allen belts and thus has a slightly higher TID 2) Cobham 65 nm component than interplanetary space [3]. The main difference 3) BAE 45 nm between the latter environments, as it pertains to mission 4) Boeing 32 nm radiation exposure is that GEO has somewhat higher TID A. Boeing 90 nm levels and generally has SEE rates between ISS and The 90 nm Boeing RHBD program is an example of a well- interplanetary space. For reference, TID rates are around 500 document RHBD library that has been used to make ASICs. rad(Si)/year in LEO and 1-10 krad(Si)/year in GEO Some of the devices made in this technology node, such as the (depending on orbit, solar cycle, and other factors). The TID MAESTRO processor had to be implemented with lower-end targets we were primarily interested in were 100 krad(Si) RHBD in some areas due to the number of SRAM bits used minimum, and a desired capability of 300 krad(Si). and the lack of space to implement the bits on the device. 90 nm is considered too large for the desired device density but is PTIONS IV. RHBD O used to establish trends. This library is considered mature, but It is possible to make standard components perform better it is also old and there is a successor in the newer 32 nm in radiation environments by altering low-level library element program from Boeing. designs (e.g. by adding in feedback loops), altering higher It is important to briefly discuss the observed memory SEE level functionality (like triple module redundancy [TMR]), or rate of 1×10-6/bit-day in ISS and 5×10-6/bit-day in solar by changing the thickness of material stacks or using alternate minimum GCR. These are a few orders of magnitude poorer materials. The collection of all these methods is referred to as performance than bits from other RHBD programs reviewed 3 later. It is believed this happened because Boeing chose to overall bit error rate lower before error correction. Also, the implement better EDAC, along with partial hardening of cells, non-RHBD version of their SRAM cells is about 10-50× better rather than full RHBD of cells. As a result, the rates for on- than the other RHBD providers. The program overview [1] orbit errors in EDAC-protected memories is still very low provides significant information on baseline library elements. (typically it can get down below 1×10-4/system-year, with This library is not expected to migrate to a smaller feature size sufficient scrubbing). The results are summarized in Table 1. in the near future. So, the parameters in the table would be The TID and SEL performance of the RHBD library expected to hold for the next 3-5 years. The logic error/SEFI developed on this fabrication line are both very good. rate shows similar sensitivity between protons and heavy ions, However, this fabrication node may be at a sweet spot for both suggesting that some SEFIs are caused by proton events SEL and TID performance. Later devices use lower voltage, (secondaries). which helps with SEL.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages7 Page
-
File Size-