Solid-state lighting (SSL) technologies and products are gradually penetrating daily life. An SSL system is composed of a light-emitting diode (LED) engine with an electronic driver or drivers, integrated in a housing that also provides optical functions, thermal management, sensing, and/or other functions. Knowledge of system reliability is crucial, not only for successful business in today’s SSL products and applications, but also to gain a deeper scientific understanding that will enable improved products and application design in the future. System malfunction may be induced by the failure and/or degradation of any subsystem or interface. Most SSL system designs, which allow few failures of the subsystem/interface during the application period, can be achieved with significant cost reduction when system reliability is well understood by means of appropriate experimental and simulation techniques.

Color maintenance is a recently-appearing system failure. Color-maintenance problems are insidious, because they are poorly understood, and only appear after many hours of operation. The purpose of this paper is to describe current practice, that is, field issues, test results, and prediction capability, with color maintenance; and to recommend how to address LED color maintenance in our products. This paper presents:

• Examples of color-maintenance issues that have appeared in the field;

• The origins of color shift;

• Current prediction insights; and

• Standardization activities.

“Color maintenance” is analogous to lumen maintenance and is defined as the change in chromaticity of a light source with respect to the chromaticity at the beginning of the lamp’s life. It is typically measured as Δxy or as ∆ u^{\prime }{v}^{\prime } in the Commission Internationale de l’Eclairage (CIE) color coordinate systems. The chromaticity coordinates of a source provide a numerical representation of the color of the light. The three most common chromaticity diagrams, with their coordinates, are the CIE 1931 (x, y), the CIE 1960 (u, v), and the CIE 1976 ( u^{\prime } , v^{\prime } ). The (x, y) coordinates are the most frequently reported. Every color is represented by unique (x, y) coordinates. The CIE system is the most common method of characterizing the composition of any color in terms of three primaries [ 1, 2]. Artificial colors, indicated by X, Y, and Z coordinates, also called tristimulus values, can be added (X + Y + Z = 1), to produce real spectral colors. The chromaticity coordinates, x, y, and z [ 1], are the ratios of the X, Y, and Z coordinates of the light to the sum of the three tristimulus values. It is necessary only to consider the quantity of two of the reference stimuli in order to define a color, since the three quantities (x, y, z) are always made to sum to 1. Thus, the (x, y) coordinates are commonly used to represent a color [ 1, 2].

The ( u^{\prime } , v^{\prime } ) coordinates are related to the (x, y) coordinates by the following equations:

Based on Eq. (1), the coordinates ∆ u^{\prime }{v}^{\prime } , which define the color shift at any two positions (0 and 1), can be calculated using the following formula:

Energy Star specifies that color maintenance must not exceed ∆ u^{\prime }{v}^{\prime } = 0.007 on the CIE u^{\prime }{v}^{\prime } diagram, after 6000 h of operation.

“Color consistency” is the variation in chromaticity at the start of a lamp’s life among a population of products. For example, a product may be made from LEDs that are binned to fall within three MacAdam steps of a target chromaticity. These LEDs have a color consistency of three steps. Color consistency can also be defined in terms of (x, y) or ( u^{\prime } , v^{\prime } ). The color consistency of lamps built from these LEDs may be worse than three steps because of temperature variations, current variations, or other factors.

“Color stability” describes how the entire spectrum changes over time. Although it is closely related to color maintenance, color stability encompasses more detail. The term “color maintenance” will be used in this paper.

Two hundred L Prize lamps were tested for 25 000 h at 45°C. Testing has continued for 32 of these lamps and test hours now exceed 36 000 h (Figure 1). The L Prize lamps show that LED technology enables color shift (and lumen depreciation) to be quite small [ 3]. The average lumen maintenance for 200 lamps is over 100%, after 25 000 h. The average lumen maintenance at 36 000 h is 96.5% with respect to the maximum light output, which occurred after about 2000 h of operation. The average color maintenance, ∆ u^{\prime }{v}^{\prime } , is slightly above 0.001, or about one MacAdam step [ 3], both at 25 000 h and 36 000 h, as shown in Figure 2. (Two light sources observed simultaneously cannot be distinguished from each other by an average viewer, if the colors are within one MacAdam step of each other.)

Controlling color shift in an L Prize lamp was particularly complicated, because two LED colors were used (red and blue) in addition to a remote phosphor. When two colors of LEDs are used, they will have different temperature dependence and different degradation rates. Both of these factors will cause color maintenance to be poorer than for a light source made from a single LED color. Still, it is possible to create a stable light source, as shown in the figures above. In some cases, color-maintenance issues appear after several years of lamp service. Because the market has the impression that LED lamps “live forever,” failures like color shift are particularly irksome.

At the US Department of Energy (DOE) R&D workshops in 2011 and 2013, Scott Rosenfeld, a curator at the Smithsonian Institution, reported color-consistency (2011) and color-maintenance (2013) problems with LED lamps [ 4, 5]. Clear color change can be observed in Figure 3. Reported color shifts at the museum were as large as ∆ u^{\prime }{v}^{\prime } = 0.027 (27 steps), after only 10 000 h of operation! Some of the reported color shifts are too large to comply with the Energy Star specification that color shift should be less than 0.007 at 6000 h of operation.

DOE-funded project at the Pacific Northwest National Laboratory (PNNL) [ 6− 8] gives an excellent overview of color maintenance. A gateway project conducted at the Smithsonian Institution, involving several lamps from six different lamp types, indicated that many of the lamps failed the Energy Star requirement that ∆ u^{\prime }{v}^{\prime } should not exceed 0.007 before 6000 h of life. The graph in Figure 4 is taken from Ref. [ 6] and summarizes the results of several Commercially Available LED Product Evaluation and Reporting (CALiPER) studies on many different products. This figure demonstrates that color shift can show quite different behavior for different lamps. A study of 17 A lamps operated for 8000 h at 45°C showed that 3 out of 8 Energy Star lamps failed to meet the 6000-hour Energy Star requirement for color shift. With such erratic behavior, one might expect difficulty in predicting color maintenance.

The CALiPER Retail Lamps Study reports on the lumen and chromaticity maintenance of LED lamps that are operated in steady-state conditions [ 7]. This report documents the long-term performance of 15 of the LED A lamps from CALiPER’s third study of lamps that are available in retail stores. Specifically, the report focuses on lumen and chromaticity maintenance relative to benchmark halogen and compact fluorescent lamps (CFLs). Talbe 1 lists the commercial lamps that were studied and tested.

The results include data on lumen maintenance for 15 commercial LED lamps, as well as the color maintenance of these LED lamps compared to conventional technologies. Figure 5 shows that from an average color-maintenance point of view, LED technology does not differ much from CFL and halogen benchmarks.

Figure 6 shows the average chromaticity maintenance for each lamp model. The three worst are the CREE, MaxLite, and Great Value lamps. These three do not meet the Energy Star specifications. The CREE lamp did not even make 1000 h before its color exceeded seven steps.

The CALiPER report also investigated the relationship between lumen maintenance and chromaticity maintenance, as shown in Figure 7. Overall, there was a correlation between lumen depreciation and color shift, but no detailed investigations toward the mechanism were presented.

This section explains encapsulant- and phosphor-related degradations.

LEDs are susceptible to contamination from several sources. A particular concern is the presence of sulfur-containing compounds, which can cause darkening of the silver mirror that is frequently used beneath LEDs to increase package optical efficiency. More details can be found in Ref. [ 21].

Figure 12, taken from Ref. [ 6], shows the effects of the delamination of a phosphor coating applied directly to the LED in a high-power package. For Figure 12(a), the phosphor has peeled up from the edges of the LED, allowing more blue light to escape and causing an overall shift toward blue. It may also cause spatial non-uniformity. Such an LED will emit more blue light to the sides and more yellow light on the optical axis. For Figure 12(b), the phosphor has lifted from the LED in the non-edge portion of the device. In this case, the average path length for a blue photon through the phosphor increases. With longer path length, the chance of capture in the phosphor increases and more blue light is absorbed in the phosphor and converted to lower wavelengths, shifting the color toward yellow.

Mid-power LED packages can also show signs of delamination. Mid-power packages frequently use combinations of metals, silicones, and epoxies that have quite different material properties (i.e., different coefficients of thermal expansion). The adherence of these materials strongly depends on the process and on environmental conditions like moisture and temperature. If any interface delaminates, the scattering internal to the package will change, and color shift is likely.

On a system level, other parts may also play an optical role by either reflecting or transmitting light, and therefore may also contribute significantly to color shift if they degrade. For example, white solder resist deposited on printed circuit boards may become yellow over time. A micro-foamed PET light-reflective sheet, or MCPET, used in diffusers, may become brittle. Over time it may show signs of cracks, which also change the diffraction of light. Secondary optics in LED sources can also degrade, causing color shift, if improper materials or thermal design are used. The lamps studied in Ref. [ 6] did not show large color shift attributable to secondary optics. Measurements before and after removal of the secondary optics did not yield large color differences, indicating that the secondary optics were not contributing to color shift. This result does not mean that all lamps have no color shift induced by degradation in the secondary optics. Care in the design of the entire optical path is required to avoid degradation of any of the elements, whether plastic optics, reflecting foils, or other components. Outdoor applications, where optics are exposed to UV light from the sun, weather, and insects, may be more susceptible to degradation of secondary optics.

Any of the degradation processes above is influenced by temperature and other stresses. Color shift will be accelerated by higher current or higher temperature. For example, Figure 13 [ 25] shows a dramatic increase in the rate of color shift as the LED junction temperature is increased. In this particular case, the color shift begins only after about 15 000 h, so lamps produced from these LEDs would only begin to show strong color shift after being installed for about two years, if operated 24 h per day. Predictions based on the first few thousand hours would reveal no such shift.

Lamps using remote phosphor are less susceptible to many of the effects described in the previous section, because delamination is not a concern, and because the remote phosphor is generally at a lower temperature than phosphors embedded in an LED package. The energy difference between blue absorbed photons and photons emitted by the phosphor is emitted as heat in the phosphor. If thermal management is insufficient in the remote phosphor, overheating and damage in the remote phosphor and its substrate can occur. Many other sources of temperature shift can occur. For example, the adhesives that bind the LED to the submount may degrade, increasing thermal resistance and LED-junction temperature, or the solder that binds the LED-package electrical and thermal contacts to a printed circuit board may fatigue and crack, reducing contact area and increasing thermal resistance and LED-junction temperature. LED color also shifts in reversible ways when the current or temperature is changed. Changes in the driving electronics that result in a different LED driving waveform (whether deliberate or as the result of degradation of the electronics) can also lead to color shift. Changes in ambient temperature generally cause reversible color shifts. Changes with temperature and current are of particular concern in light sources that contain more than one LED color, because the different LED colors are likely to have a different dependence on temperature and current. In this case, even though the color of the individual LEDs may shift insignificantly with the temperature/current change, the fact that the amount of light from the different LEDs changes differently can result in much larger color shifts in the combined light.

LM-80 is a method for measuring lumen depreciation and maintenance developed by the Illumination Engineering Society (IES) to enable comparison between LEDs from different manufacturers. LM-80 data includes color measurement. Thus, any LED qualified for use in Energy Star products will have color maintenance data available for at least 6000 h. This data may be helpful in determining if a particular LED package is likely to have a color maintenance issue. Figure 14 plots LM-80 data for two typical high-power packages at 85°C. The color maintenance looks reasonable at currents less than 1000 mA. However, drastic color shift is apparent at 1000 mA. Can LM-80 data like this be used to predict color-maintenance problems? Data in Figure 14 extends only to 85°C, which is a relatively low temperature, and may be a warning sign in itself (other LED manufacturers report data to 105°C, 120°C or higher) [ 8]. Data at higher temperatures might have revealed the color-maintenance problem with these LEDs more clearly.

In addition to thermal effects, the main reason for the discoloration and yellowing of optical materials within LED-based systems is continued exposure to wavelength emission (blue/UV radiation). Photo-degradation of the polymer materials used for LEDs usually takes place as a result of increasing the molecular mobility of the polymer and/or introducing chromospheres as an additive into the molecule. Both changes result in absorption maxima in a region where the matrix polymer has no absorption band [ 25]. Photo-degradation also depends on exposure time and the amount of radiation. The chemistry of degradation processes in polycarbonates has been studied extensively over the past few decades [ 26− 28]. In BPA-PC, the chemistry underlying photo-degradation has been described in two different mechanisms: the photo-Fries rearrangement and photo-oxidation. The relative importance of these two mechanisms depends on the irradiation wavelengths applied. Previous investigations show that the photo-Fries rearrangement reaction is more likely to occur at wavelengths shorter than 300 nm, whereas photo-oxidation reactions are more important when light of longer wavelengths (>340 nm) is used [ 26− 28]. When light with wavelengths longer than 340 nm is used, the dominant photo-degradation reaction is reported to be side chain oxidation [ 27].

In addition to the effects of light, the yellowing and degradation of package materials is largely dependent on temperature, which is a combination of junction temperature, ambient temperature, and LED self-heating [ 9]. Narendran et al. [ 11] reported that degradation was affected by junction heat and the amount of short wavelength emissions. It was shown that the thermal effect has a greater influence on yellowing than short-wavelength radiation. Figure 15 depicts the color shift (∆ u^{\prime }{v}^{\prime } ) of the specimens at different loading conditions. There is no major color shift during the incubation stage, whereas the color shift during the degradation stage is linearly proportional to the testing duration [ 22].

The yellowing index (YI), which shows the amount of discoloration, of BPA-PC plates was also measured as a function of exposure time under thermal and thermal plus blue-light stress, as shown in Figure 16. This figure shows the YI of the sample, exposed to 140°C without irradiation of blue light, together with the YI of the photo-aged sample (at 140°C). Obviously, the temperature itself could cause the yellowing. However, one can see that the blue light has a significant contribution to the yellowing. This aging test was done in almost pure BPA-PC, which is used as a lens and as a substrate for phosphor in remote-phosphor LEDs, in order to clarify the chemical reason for color changing in BPA-PC; details of the test are already published [ 19]. In commercial samples, light has a lesser effect on discoloration due to the use of heat stabilizers, and the main reason for yellowing in these samples is thermal aging. For this reason, thermal aging is used to accelerate degradation mechanisms in order to study color shift. It is already reported that by increasing the temperature, the rate of discoloration is faster [ 19− 23]. In our next work, the effect of light on the aging of remote phosphor and color shift, and possible solutions for reducing the rate of color shift, will be studied extensively.

Thermal aging at 85°C for up to 3000 h was also done for PMMA, which is used as a secondary optics. It has been shown that there is no significant aging for PMMA materials under any implemented aging stress for up to 3000 h at 85°C [ 28].