Bill of Attributes, Life Cycle Assessment and Materials Flows:
Case Study of Laptop Computers
Eric Williams*1, Callie Babbitt1, Ramzy Kahhat2, Barbara Kasulaitis1
1
Golisano Institute for Sustainability, Rochester Institute of Technology, Rochester, New York, USA 2
Department of Engineering, Pontifical Catholic University of Peru, Lima, Peru
* Corresponding Author, [email protected], 1-585-475-7211
Abstract
Life Cycle Assessment (LCA) and Material Flows Analysis (MFA) both critically depend on the attributes of products such as content of materials and parts (Bill of Attributes = BOA). While there has been significant work to develop commercial and public databases detailing production processes, there have been no comparable ef-forts to characterize BOA. Two issues that could critically affect LCA and MFA results are variability in BOA within a product class and evolution of BOA over time. This study examines the above issues through a case study for laptop computers, involving disassembly, BOA characterization and analysis for seven models of differ-ent years, makers and screen size. While the limited sample size does not permit conclusive characterization of trends, initial results suggest significant variability in material content and attributes among models. The silicon wafer area associated with memory appears to have increased dramatically from 1998-2008.
1
Introduction
The life cycle inventory for manufacturing a product combines information about a product’s attributes with supply chain data on materials use and emissions in processes. Mathematically this relationship can be written (in this example for carbon dioxide) as the dot product:
Life cycle carbon = BOA · CO2/attribute (1)
where BOA (Bill of Attributes) is a vector of product attributes such as masses of materials, content of components and other information and CO2/attribute is a vector of supply chain emissions to deliver a unit of attribute (e.g. CO2 emitted per kg of aluminium). BOA has been long neglected in the LCA develop-ment community. While there has been significant work to develop both commercial and public process databases, there have been no comparable efforts to characterize BOA. The implicit assumption is pre-sumably that since the product under study is “in hand” and processes are “out there”, analysts can ob-tain their own BOA and the need for support is on the process side. This logic is sound in principle but fails in practice. For complex products in particular, and indeed most products in the modern economy are complex, disassembly is labor intensive and reverse engineering internal attributes such as materials con-tent requires sophisticated equipment such as a mass spectrometer. Further, building BOA through gather-ing information throughout the supply chain is possi-ble in principle but faces many of the same challenges
with gathering process data, including proprietary is-sues.
From equation (1) is it clear that reliable BOA data is as important as reliable process data in the carbon footprinting of products. As with process data there are important methodological questions to address for BOA. What kind of data needs to be gathered? Are there useful units of aggregations of product attributes that can streamline data gathering yet still lead to reli-able footprint results? When can secondary data be used versus as opposed to collecting primary data for a product?
BOA is also important for Materials Flow Analysis (MFA). MFA generally aims at estimating more ag-gregated flows compared to often product specific LCA. Variability of materials content within a product class and over time can significant affects MFA re-sults.
The appropriate data structures and collections meth-ods for BOA vary by product. Some products such as detergents the BOA changes slowly while for sectors such as information technology products both BOA and process material flows can be rapidly moving tar-gets. The techniques and labor needed to determine BOA also vary.
This article aims to contribute to developing data and understanding of the BOA vector of the equation for laptop computers. The broad goals of this study are: • Generate primary data on BOA through physical dis-assembly and x-ray analysis,
• Characterize trends and relationships between differ-ent attributes,
• Explore different characterizations of product attrib-utes (e.g. silicon wafer area versus packaging area) for use in LCA.
2
Methods -Analytical
BOA can take many different forms but in general it can be broken into subcomponent vectors:
BOA = function (materials, components, assemblies) (2)
These sub-component vectors are not independent, for example, the content of silicon chips in the component affects the quantities of gold in the materials compo-nent. To discuss each component sub-vector in more detail, the materials piece is the most conceptually simple. It can be written as:
materials = (mass of constituent material 1, mass of constituent material 2, …) (3)
where constituent materials are substances such as steel, aluminium, different types of plastics and pre-cious metals. Some of these materials (e.g. steel) are relatively easy to separate and identify with disassem-bly while other materials, such as precious metals, re-quire special processing to measure quantities or data from suppliers. Note that while [3] looks simple, the situation is complicated by different grades and com-binations of materials. A more disaggregated descrip-tion pulling out different grades and combinadescrip-tions of materials could make a significant difference in the carbon footprint, for example, the purification of in-dustrial grade materials to semiconductor production grade. Note that in the formalism one could choose to put materials processing in the materials, e.g. a sepa-rate category for purified silicon. Alternatively,silicon processing could be included in the integrated circuit piece of the BOA vector.
Note that breaking down laptops into constituent ma-terials is labor intensive and for some may require special laboratory equipment. Rather than undertake a materials analysis for each and every analysed laptop, it is clearly desirable to find regular relationships be-tween materials contents and other product character-istics. Given that the total mass of laptops varies sub-stantially year-to-year and model to model, one ave-nue to explore relationships is to express the mass vector as a total mass and set of mass fractions:
mass =
(total mass, mass fraction 1, mass fraction 2, ….) (2) There are many choices for how to define bills of at-tribute and combine with process data. The central
is-sue is that the two must match. For example, if the process data available for semiconductor manufactur-ing is in the form of carbon dioxide per area of silicon wafer area, in turn the BOM definition in components must reflect contained wafer area. The lack of disag-gregated process data is a central driver of what com-ponent definitions are feasible. For example, while it would presumably be more accurate to describe inte-grated chip content as a list of model numbers, this would not be useful unless process data for different chip models were available. Most available process data for electronics components is highly aggregated, e.g. in the form of energy use to manufacture an ag-gregate mix of different types of chips, circuit boards, discrete components, etc. Until more disaggregated process data becomes available in most cases the BOA data complementing the process data will be aggre-gated.
It is important to identify patterns and relationships that both simplify the collection of BOA data and clar-ify what forms of BOA data are most important for reliable LCA. From these goals two sub-questions emerge:
• Are there regular relationships between mate-rial and component characteristics and macro product attributes? For example, does silicon wafer content correlate with screen size of a laptop computer? • Can easily visible component characteristics be mapped to ones more difficult to measure? For ex-ample, does the packaging area of a chip correlate with contained silicon wafer area?
There are clearly many possible relationships between product attributes one could test. The goal of this study is begin the process of developing BOA meth-ods through analysis of a small subset of possible rela-tionships. With luck useful relationships will be found, but since it feeds into future work, even failure of a hypothesized relationship provides useful information. At the least this effort will help clarify how to proceed with the BOA.
3
Laptop Disassembly
We disassembled seven laptops (12”, 14.1”, and 15” screen sizes) manufactured in different years, details of the models processed are shown in Table 1. The masses of bulk materials, areas of motherboards, and characteristics of chips on motherboards (number, packaging area, and silicon wafer area) were meas-ured.
The disassembly process begins with the identification and disassembly of the laptop into major component groups. The groups were chosen based on ease of dis-assembly and functionality. Each component group
represents one assembly that performs a specific func-tion in the laptop. Components include hard disk, DVD-CD drive, floppy disk drive, battery, mother-board, modem module, LCD display, computer chas-sis, and others components. Power adapters have been excluded in this disassembly process. After identifica-tion and basic disassembly, each component group was then weighed and sorted for further disassembly and detailed inventory. Once the major components were separated, each component group was then fur-ther dismantled into individual components to be sort-ed by material type, weighsort-ed and catalogusort-ed. The dis-assembly was done using hand tools, including screwdrivers, pliers and wire cutters. Each part was completely disassembled to the level where each piece comprised of a single material. Disassembly was car-ried out until all materials were separated or no further mechanical separation was possible. Common materi-als were grouped together for cataloguing if they con-tained a number of very small and similar parts, such as screws, wires and adhesive tape. Larger pieces were identified and weighed as a single piece. Some com-ponent groups contained more than 30 parts when dis-assembled. Each part was given a unique part number for the ease of recording and identification. Material types for each part were then identified by physical inspection and categorized into generalized groups. For example, all types of aluminium are considered equal. Different alloys of aluminium were not identi-fied. A magnet was used to separate ferrous and non-ferrous metals. The items classified as magnesium were identified by an Mg symbol placed by the manu-facturer. It is important to note that some components were comprised of multiple material types, and physi-cal separation could not be accomplished. When these circumstances were encountered, the ratio of material types based on weight was estimated
The silicon wafer area for the motherboard and mem-ory cards for the 2008 Hewlett-Packard Elitebook 6930p was measured using a Glenbrook JewelBox 70T X-Ray Inspection System and associated GTI-5000 software. The circuit board was positioned in the X-ray Inspection System and the software was cali-brated for the height of the manipulator arm. An exist-ing hole in the motherboard was used as the reference to calibrate the software. All silicon wafers on the top side of the board were measured and photographed. The board was then turned over to complete the meas-urements of the silicon wafers on the bottom side. Prior to completing the measurements on the bottom side of the board, the calibration was verified using the reference hole. In cases where two components were adjacent on opposite sides of the board, photo-graphs of the circuit board were used to clarify the identity of the components. To complete the
meas-urements for the memory card, the height of the ma-nipulator arm was changed to accommodate the larger wafer area, and the software was recalibrated using a reference hole on the memory card. Because the size of the wafer area on the memory card required the X-ray system to be used at its lowest height and maxi-mum zoom, a wafer on the memory card was also ground and measured to verify the measurements taken with the machine. Several measurements were also taken with another circuit board that had been ground and measured, in order to verify the accuracy of the measurements taken with the X-ray system. Table 1: Laptop models disassembled. All models had internal CD-ROM or CD-ROM/DVD drives. (P=Pentium)
4
Results
4.1
Silicon wafer area
Energy use and other environmental impacts associ-ated with semiconductor manufacturing and associassoci-ated supply chains can make significant contributions to life cycle totals for a computer [1]. The impacts of semiconductor manufacturing is thought to be domi-nated by fabrication of microcircuits on silicon wafers [2] wafer area is often used in the industry as a
nor-# Year Model Scrn Size Wgt. (kg) D-RAM (MB) CPU Hard Drive (GB) 1 199 9 Dell Lat-itude R-Series PPX 14.1 2.99 64 P II 400 MH z 6.4 2 200 1 Dell Inspiron 4100 14.1 2.74 128 P III 1.2 GHz 20 3 200 2 Dell In-spiron 2500 15 3.66 512 Mo-bile P III 10 4 200 3 Dell In-spiron 5100 14.1 3.26 256 P 4 2.4 GHz 30 5 200 4 Latitude D600 14.1 2.44 512 P M 1.40 GHz 30 6 200 8 HP Elite Book 6930P 14.1 2.41 4096 Core 2 Duo 2.80 GHz 160 7 200 8 HP Elite Book 2530P 12.1 1.74 3072 Core 2 Duo 80
malization unit for energy and materials flows. Figure 1 shows the measured silicon wafer contained in the CPU, DRAM and Motherboard (MB), the latter figure exclusive of CPU and DRAM area. The most notable trend in the rapid increases in silicon area associated with DRAM. The degree to which this rapid growth in DRAM silicon wafer imply increases environmental burdens depends on the evolution of manufacturing processes. Aggregate energy use per area of wafer fab-ricated has shown steady, though slow, decreases [3]. Figure 1: Trends in silicon area in Motherboard, CPU and DRAM. Summary information on models:
1: 1999, 14.1” screen, 64MB DRAM 2: 2001, 14.1” screen, 128 MB DRAM 3: 2002, 15” screen, 512 MB DRAM 4: 2003, 14.1” screen, 256 MB DRAM 5: 2004, 14.1” screen, 512 MB DRAM 6: 2008, 14.1” screen, 4096 MB DRAM 7: 2008, 12.1” screen, 3072 MB DRAM
Figure 2 shows how silicon wafer area per megabyte of DRAM varied over the different models. Not sur-prisingly, there is a pattern of decreasing silicon area over time. The pattern discernable from Figures 1 and 2 is that decreasing feature size in semiconductor manufacturing enabled more memory to fit in same area, but growth in demand for memory in personal computing far outstripped this progress.
Figure 2: Ratio of silicon wafer area to megabytes of memory in different models
4.2
Relationship between silicon wafer
and packaging area
While the use of the X-ray inspection system simpli-fies data collection for silicon wafer contained in mi-crochips compared to grinding chips, it would be ideal if one could estimate the environmental impacts from externally measurable characteristics such as the area of packaging or number of pins. The hope is a repro-ducible relationship exists between internal and exter-nal characteristics. Figure 3 explores the relationship between the area of contained silicon wafer area and external packaging, plotted as a function of the year of manufacture. There are observable temporal trends, roughly stable ratio for motherboard chips (in the ag-gregate) and a rapidly increasing ratio for memory chips. Any mapping between silicon and packaging of memory, for example, must account carefully for year of manufacture. There is also variability between models that depending on the intended accuracy of a LCA, could affect results if packaging area were used as a measure.
Figure 3: Silicon wafer area / Chip packaging area for different classes of microchips in laptop computer
4.3
Bulk Materials
Figure 4 shows the fractions of different bulk materi-als identified in the seven models. Structural materimateri-als (aluminium, steel, plastic, and magnesium) represent around half the laptop weight, the mixes of these four materials varies model by model. The two 2008 mod-els show a substitute from plastic to magnesium, but given the mix of business versus home models and manufacturers, it is not clear is this is a general tempo-ral trend for laptops. The significant share of magne-sium does imply a shift in the number and value of bulk materials. We noted that the magnesium pieces in the two 2008 laptops were glued to other frame mate-rials and thus difficult to manually recover.
0 500 1,000 1,500 2,000 2,500 3,000 1 2 3 4 5 6 7 S il ic o n a re a (s q u a re m il li m e te rs ) Model Number Motherboard CPU DRAM 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 1 2 3 4 5 6 7 W a fe r a re a to m e m o ry ra o (m m 2 / M B ) 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1999 2000 2001 2002 2003 2004 2005 2006 20072008 R a o o f si li co n w a fe r a re a to p a ck a g in g a re a Year MotherboardIC CPU Memory
Linear(MotherboardIC) Linear(Memory)
Figure 4: Shares of material weights in laptop com-puters
5
Discussion
Results are subject to the caveat that the sample of laptops analysed mixed characteristics, years and manufacturer in insufficient numbers to accurately separate the effects of different variables.
We found large variations in BOA for laptop com-puters, the most striking of which is an order of mag-nitude shift in silicon content of DRAM over time. Whether or not such variability would influence quali-tative conclusions from an LCA or MFA would de-pend on the analysis. The potential for variability, par-ticular that driven by technological change, appears significant enough to call for inclusion of BOA in fu-ture efforts to analyse ICT products.
The parameterization of attributes in order map to process data is non-trivial. Different choices to pa-rameterize semiconductors, for example, such as chip silicon area, packaging area, mass, and pin number could change LCA results. The appropriate matching of attribute choice and process data to improve accu-racy is an important issues that has yet to be explored. While this work focuses on laptop computers, the is-sues raised here are general. Variability of BOA for a product class can be significant, affecting the reliabil-ity of LCA and MFA. Variabilreliabil-ity and uncertainty in BOA, to our knowledge, has yet to be considered however.
6
Literature
[1] L. Deng, C. Babbitt, and E. Williams, “Eco-nomic-Balance Hybrid LCA Extended with Un-certainty Analysis: Case Study of Laptop Com-puter”, Journal of Cleaner Production 19(11): 1198-1206 (2011)
[2] E. Williams, R. Ayres, and M. Heller, “The 1.7 kg microchip: energy and chemical use in the pro-duction of semiconductors”, Environmental Sci-ence & Technology 36 (24), 5504-5510, Dec. 15 (2002)
[3] L. Deng and E. Williams, “Functionality versus “Typical Product” Measures of Energy Effi-ciency: Case study of Semiconductor Manufac-turing”, Journal of Industrial Ecology 15 (1) : 108–121 (2011) 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1 2 3 4 5 6 7 W e ig h t sh a re o f m a te ri a l Model Number Others LCDMaterials PWBMaterial Ba eryCell OtherPlas c PC+ABS Ferrous Copper Magnesium Aluminum
