FoodSkin: Fabricating Edible Gold Leaf Circuits on Food Surfaces

We present FoodSkin, a technique for adding interactive elements to foods by implementing edible circuits on the surface of the food. The circuit is easily fabricated using commercially available materials. Existing approaches to enhance the eating experience, such as presenting an electrical taste by making food part of an electronic circuit, are challenging to apply to foods with low water content due to their low conductivity. Our technique enables the integration of dry foods into an electronic circuit and provides displaying (e.g., smell or taste) and sensing (e.g., eating activity) functionalities. We describe our fabrication technique with a library of food materials that we can utilize, evaluate the conductivity and adhesion of the gold-leaf traces, introduce demonstrative applications, and conclude with a workshop we conducted to evaluate the accessibility of our technique. FoodSkin enriches the design space for the computer-augmented eating experience by enabling the digital fabrication of electronics on versatile materials, surfaces, and shapes of foods.


INTRODUCTION
Food plays an essential role in human life, serving as more than a source of nutrition: it also provides pleasure through its favors, textures, and visual presentation.In addition to culinary enthusiasts and professional chefs, the search for novel culinary techniques and unique ingredient combinations has attracted the attention of manufacturers and researchers [2,46,66].Recent explorations into Human-Computer Interaction (HCI) have led to a greater convergence between food and computing, resulting in an emerging feld of research known as Human-Food Interaction (HFI) [3,4,17].In this domain, researchers have proposed various approaches to enhance the eating experience.Several of these focus on visual enhancements, such as projecting images onto a dining table for appealing food presentation [18] or using augmented reality to alter the appearance of food and infuence perceptions, including satiety [59].In addition, researchers have proposed techniques to incorporate functionalities into food [16,42,43,62].For example, they have explored methods to control satiety [42], have added passive tags [52] by altering the internal flling structure of 3Dprinted cookies (using diferent shapes of 3D-printed chocolates to achieve cross-modal efects and control taste variation [62]), and have added a shape-changing functionality to foods [61,72].
More recently, an emerging stream of research has focused on enabling food itself to function as an interactive electronic device that provides sensing or display capabilities [31,33,56,74,76].These techniques have been efectively applied in various contexts, enabling dynamic electric taste presentations [56], controlling taste changes over time in beverages [74], displaying information on food [31] and drink surfaces [32], and ofering interfaces for food education [33] or enhancing social communication [76] through auditory feedback from food consumption.However, many of these techniques rely on the food's conductivity, using the food itself as part of an electronic circuit.This confnes their application mainly to high-moisture foods [31,33,56,74,76].Dry foods have low conductivity, making it challenging to apply these techniques, which signifcantly limits the design opportunities for HFI and prevents these techniques from being adopted in a wider range of applications.
We propose FoodSkin, a technique that uses edible gold leaf to fabricate electronic circuits on dry food surfaces, thus enabling the integration of input and output (I/O) functionalities into various foods (Fig. 1).Edible gold leaf (24K gold), which is tasteless and odorless, serves as a food additive that does not signifcantly alter the original taste or favor of the food.Moreover, it has a long history of use as a decorative material due to its beautiful luster.More recently, its aesthetics and conductivity have led to it being applied to the surfaces of various objects to create electronic circuits [36,67,69].The FoodSkin fabrication process is simple, requiring only readily available materials (i.e., gold leaf, edible paper, and potato starch) and low-cost personal fabrication tools such as laser cutters.This approach enables the integration of dry foods with electronic circuits.
Our technique aims to (1) broaden the scope of designing existing food interactions that rely on conductivity to encompass dried foods, and (2) provide novel interactive experiences with food by incorporating circuits into it.Concerning point (1), we utilize our proposed technique, targeting dried foods, to replicate the interactions proposed in previous work.Specifcally, this involves changing favors through electric stimulation of the oral cavity [55] and ofering feedback when the food touches the oral cavity or tongue [33,76].Regarding point (2), our technique, which allows circuits to be directly embedded in food surfaces, has the potential to signifcantly expand the design possibilities for enhancing the computer-augmented eating experience (Fig. 1 (c)-(e)).This technique is the frst to incorporate localized heating and olfactory enhancer features on dry food surfaces.In this paper, we go through the details of how to augment food with the FoodSkin technique, considering its material properties and design support system.Our main contributions are summarized as follows.
• Presenting a technique for creating gold leaf-based circuits on food surfaces.• Investigating the conductivity and adhesion of gold-leaf traces on various foods, and utilize the subsequent insights to develop a design support system.
• Evaluating the impact of our fabrication process on the original taste and texture of food through a user study.• Demonstrating several applications of FoodSkin that showcase its diverse functionalities.• Conducting a workshop to evaluate the accessibility and feasibility of our technique.

RELATED WORK 2.1 Fabrication of Customized Edible Electronics
In the feld of HCI, researchers have explored various approaches for fabricating circuits on objects, including techniques such as inkjet printing [40], screen printing [15], spraying [78], painting [12,77,84], and water transfer [22], which involve the use of conductive ink.Other approaches involve 3D printing with conductive flaments [29,38], attaching conductive tape [37,44,68,81], and weaving conductive thread [70].There has been growing interest in utilizing edible materials for circuitry, leading to the development of edible electronic circuits [14,24,30,39,64,79].These circuits are designed to be compatible with digestion in the body, holding promise for applications in healthcare and medical monitoring [9,79].Various materials have been utilized as edible conductive materials thus far, including yeast extract spreads [24], gold leaf [39], squid ink [20], activated charcoal [79], and carbonized foods [79].Among these, gold leaf stands out due to its low resistance, lack of taste or odor, and minimal impact on the original favor of the food.For example, EdiSensor [64] uses a sheet of gold leaf on the food's surface as an electrode to detect chewing activities within the user's mouth.However, a challenge with gold leaf circuit fabrication is that the gold leaf is thin and delicate, making it difcult to create arbitrarily shaped traces [36].Although researchers have investigated techniques to create such traces using tattoo paper [68] or toner [69], the methods involved non-edible materials in the manufacturing process.
More recently, several techniques have been proposed for shaping gold leaf into arbitrary forms without using non-edible materials [14,39].In these methods, edible paper serves as a backing for the gold leaf, and a laser cutter is utilized to form the desired traces.In our previous technique [39], the edible paper remains only in the patterned area of the gold leaf, while in Cheng et al. 's [14], the edible paper covers the entire surface where the gold leaf is applied.Although edible paper left on the food's surface is tasteless and odorless, when it comes between the food and the tongue, it acts as a barrier and hinders the perception of the food's taste.In addition, Cheng et al. 's technique [14] primarily focuses on 3D-printed chocolate, which limits its adaptability to certain foods (i.e., having a melting nature when exposed to heat).We improved our previous technique [39] to enable us to create higher resolution traces than the original one.We also conducted a thorough investigation of FoodSkin to assess its durability, adaptability to foods with various surface characteristics, perceived taste and texture, and ease of handling during processing.

Human-Food Interactions
In the discipline of HFI, various techniques have been proposed to enable interactions between humans and their food.Focusing Table 1: Approaches for providing sensing and display functionality for HFI.

HFI Approaches with Food Processing.
Researchers have proposed interaction techniques that afect human perception (e.g., sight, taste, and tactile sensation) by altering the physical structure or chemical composition of foods through processing.These techniques often involve processing with personal fabrication devices (e.g., 3D printers) to form food into various shapes by layering edible materials [82], or to modify the texture of food by changing its internal structure [42,43].Another approach involves using a laser cutter to engrave information on food surfaces [21].Several techniques utilize pH-reactive edible materials to incorporate shapechanging functionality into the food [75], or to display information on the food's surface [31].According to Ogata et al. [62], changing the appearance of food afects not only the visual enjoyment but also the perception of taste.It is also possible to present taste by combining a taste solution with visual information [47,50].
Researchers have also proposed techniques to integrate sensing capabilities directly onto food by using conductive edible materials and forming electrodes [26,39,64] or antenna patterns [14].In these techniques, the systems can detect user behavior through foods, such as when the user touches the food with a tongue or fnger, the amount of food consumed, and how often the user bites into the food.

HFI
Approaches without Food Processing.Techniques have also been proposed to enable display capabilities by directly controlling various human perceptions while eating, rather than processing the food in advance.For example, while a user is eating, the taste of food can be altered through cross-modal efects by providing visual information [58] and combining visual and olfactory information [59], and the texture can be altered through auditory information [41].Techniques have also been developed to control the sense of taste itself.These include approaches for stimulating taste-sensing nerves and cells with electrical stimulation using tongue-clamping devices that directly contact electrodes [65], tableware-type devices [60], and devices that attach electrodes to the skin for stimulation [54].These techniques enable us to stimulate taste receptors and to control the taste we perceive by transmitting chemical substances in food.
Several techniques have been proposed to sense the user's eating behavior without processing the food itself.For example, to estimate what a user is consuming and how much, Mirtchouk et al. [48] used multimodal sensing (in-ear audio and motion), and Zhang et al. [83] utilized data from ambient light, lean-forward angles, and energy signals for the detection of chewing sequences.Other approaches use the detection of a closed circuit created through the food or the detection of a change in the electrical resistance of the food when connected to the device [34,76].These techniques have been integrated into devices intended for eating activities (such as tableware-type devices [34]), and are also compatible with wearable devices (such as glove-type devices [35]) that can be used during meals.In addition, these methods can provide auditory or visual feedback based on the relevant interactions [34,76].

2.2.3
Role of Food's Conductivity in Electrical Circuit.Whether or not a food is conductive will afect the functioning of several of the techniques discussed above.Specifcally, if food is included as part of the electrical circuit of the system [26, 33-35, 54-56, 76], it will work correctly if the food is conductive, but not if it is not conductive or has very high resistance values.While some of these techniques utilize a constant-current circuit, this circuit cannot apply a voltage higher than its own supply voltage.In addition, the use of a constant-current circuit for high-resistance materials can be limited by the circuit's ability to safely and efciently manage heat dissipation and power losses.Thus, it is impractical and potentially unsafe to use a constant-current circuit for foods with very high resistance values.
The technique introduced above for utilizing conductive foods as part of a circuit [26, 33-35, 55, 76] can be applied to dried foods by using our technique.In this paper, we successfully replicated these techniques with dried food (see details in Sec. 6).However, in the case of transcutaneous electrical nerve stimulation techniques [6,54], our technique cannot be applied because saliva in the oral cavity moistens the food, thereby turning even initially dry food into an electrical conductor through this moistening process.It is also difcult for our technique to trigger ionic migration [6,56].This is because the small distance between the tongue and the electrode and the small amount of electrolyte present between them make it difcult to induce sufcient migration.

DESIGN AND FABRICATION
We explain the design (Sec.3.1) and fabrication (Sec.3.2) workfow of FoodSkin.The workfow starts with designing the gold-leaf traces, followed by processing the gold leaf using the output fles generated by the design support system and connecting it to external components such as microcontrollers.We illustrate all the steps using the example of making a resistive heating circuit with a comb-shaped trace.This example is also used to introduce details of FoodSkin in our workshop, as described in Sec. 7.

Design Support System
We implemented a design support system using Java to design goldleaf traces for adding interactive elements into food and outputting fabrication fles for the subsequent processes.
3.1.1Selecting 3D Model.First, as shown in Fig. 2 (1), users select a 3D model of the food from a preset menu.When selecting basic shapes such as a box or cylinder from a preset, the users can specify the food dimensions.As an alternative method, we also implemented the functionality to import model fles (.obj fles).The fle import feature is intended mainly for 3D scanning of targets from personal devices such as LiDAR-equipped iPhones.For our prototyping in this paper, we used Scaniverse1 , an iOS app for 3D scanning.

Selecting Interactive Element.
Users select the interactive elements to be added to the food from the library, and the system opens the 2D editor for the gold-leaf trace and the 3D model viewer to preview the gold-leaf traces applied to the surface of the selected 3D model (Fig. 2 (2)).Our library supports the following interactive elements.
Heating.By creating a meandering, comb-like trace, a heater can be integrated into the food.When voltage is applied, the trace produces sufcient heat to be perceived by humans (Fig. 3 (a)).Electrical taste.The output of the constant-current circuit is transmitted to the oral cavity via gold-leaf traces, enabling electric taste (Fig. 3 (b)).Electric current applied to the tongue from an arbitrarily shaped gold leaf electrode (e.g., circular or mesh) passes through the human body and fows from the electrodes placed on the skin to the ground.The electric current applied to the tongue can present an electric taste to humans by stimulating receptors in the tongue [55].Capacitive sensing.Touch sensing can be achieved using arbitrarily shaped traces with the same mechanism as a conventional capacitive touch sensor (Fig. 3 (c)).When parts of the human body (e.g., a fnger or tongue) approach these traces, a change in capacitance facilitates touch detection.Resistive sensing.Resistive sensing is achieved by using meshshaped traces.As food is consumed and traces are partially broken,   the resistance increases, enabling the loss of traces to be estimated (Fig. 3 (d)).

Editing Gold-Leaf Traces.
Next, users specify the area where the interactive element is to be applied to the selected 3D model, and the system previews the preset traces of each interactive element (in this example, traces for a resistive heating circuit) to ft that area.Users can edit traces (e.g., reposition, remove, or add) through the editor.Using the Clothing Manipulation algorithm [28], we implemented a functionality where the trace area moves along the 3D model surface during the dragging operation (Fig. 2 (3)).During this operation, the system compares the curvature of the 3D model with the data on applicable curvatures from the technical evaluation in Sec. 4. If the trace is dragged to a position with a curvature outside the acceptable range, a warning is displayed.This estimation is based on the technical evaluation in Sec. 4. The deformation of the traces by this process is also refected in the fnal output fabrication fle, thus eliminating the need for tedious manual adjustments for shapes and positions of traces to match the model surfaces.
In this example, the preset traces consist of a comb-shaped wire and circular connectors.When users edit the area where the traces are applied to ensure that the trace acts as an electrical heating wire, the system automatically adjusts the width of the comb-shaped wire to maintain a constant resistance value (approximately 80 Ω/sq).Several other assistive features, such as automatic trace modifcation or a warning indication to avoid short circuits, have also been implemented to enable even novices with limited knowledge of electronics to design interactive elements in a simplifed manner.

Generating Fabrication Files.
Once the user is satisfed with the design and clicks on the export button, the fabrication fles in the PDF format containing vector data of traces will be generated as output (Fig. 2 (4)).

Fabrication Technique
We propose a fabrication technique that involves applying gold leaf to food surfaces by cutting them into arbitrary-shaped traces after reinforcing the gold leaf with edible paper.

Materials for Fabrication.
We used commercially available gold leaf (cost of 0.27 /2 , thickness of 0.12 , and surface resistance of approximately 0.7 Ω/sq), edible paper, and wax-coated paper 2 , which are edible or food-safe materials approved under food hygiene regulations of Japan.
After an extensive review of diferent types of edible paper (such as rice paper and sugar sheet), we found that only wafer paper met our requirements, i.e., tasteless, odorless, thin, and commercially available in our country.Wafer paper, primarily composed of starch, is mainly utilized for pharmaceutical purposes (approximately 0.01 mm thickness) 3 , confectionery packaging (approximately 0.03 mm) 4 , and food decoration (approximately 0.20 mm) 5 .The wafer paper with a thickness greater than 0.04 mm but less than 0.2 mm was not commercially available in our region, possibly due to its limited anticipated applications.We will provide a detailed comparison by thickness in Secs.4, 5, and 7. Any of these thicknesses of wafer paper can be utilized in our fabrication technique.

Fabrication
Process.The fabrication process comprises six steps (Fig. 4), as follows.(1) Position the wafer paper over the gold leaf and then cover both with wax-coated paper.(2) Using a soft material (such as a sponge or a brush), gently rub the top surface of the wax-coated paper.Then, fip it over and rub the entire back surface as well, resulting in the adherence of the gold leaf and wafer paper through static electricity.Note that gold leaf, due to its delicacy and thinness, is challenging to cut with a laser cutter or handle with tweezers, as it can easily be damaged.Therefore, we apply wafer paper to the back of the gold leaf for added support.
(3) Place two layers of wax-coated paper on top of each other and secure them to the cutting bed of the laser cutter.(4) Using the laser cutter 6 , perform cutting processes with the vector data generated from the design support system on the wax-coated paper prepared in Step 3 to create a trace.(5) Insert the gold leaf (adhered to the wafer paper in Step 2) between the wax-coated paper used in Step 4 and perform the cutting process similar to Step 4. ( 6) Apply a thin adhesive layer (potato starch dissolved in water and heated to thicken, 2.9-3.3 wt%) to the surface of the food using a paintbrush.Carefully place the gold-leaf trace on the surface of the food with tweezers so that the wafer paper side is in contact with the adhesive layer.Once the adhesive layer is dry, the gold-leaf trace functions as an electrode or wires.
This technique was based on our previous work [39].However, their approach had limitations as some portions of the wafer paper melted and adhered to the wax-coated paper during the laser cutting process, potentially damaging the gold leaf upon peeling.Specifcally, it was challenging to create traces with line widths of 2 mm or less while maintaining fabrication quality.In general, the ideal approach for circuit fabrication should be capable of high resolution, providing thin traces to facilitate intricate circuitry patterns [13].Our envisioned interactive elements also require thin traces (e.g., resistive sensing or resistive heaters).In our technique, which includes a simple but crucial refnement, specifcally in Steps (3)-( 5), we frst create hollow spaces corresponding to the shape of the trace on the two-layered wax-coated papers.The gold leaf and the wafer paper are then sandwiched between the wax-coated paper sheets before cutting, thereby reducing the adhesion between the wafer paper and the wax-coated paper and preventing damage to the gold leaf's conductive trace during removal and due to exhaust wind pressure.

TECHNICAL EVALUATION
We carried out technical evaluations to better understand the adhesion process and to assess the electrical and mechanical performance of our technique.We focused on two key factors: (1) ensuring the adhesion of gold leaf traces to substrates, and (2) maintaining their conductivity post-adhesion.Regarding (1), the adhesion possibility of the gold leaf traces depends on the surface characteristics of the substrate (e.g., water repellency and water absorbency, roughness), so testing on surfaces with diferent properties is crucial.To check for peeling, we inspected the traces after the adhesive layer dried, using both visual and tactile inspections.Regarding (2), although the gold leaf is highly conductive in its unprocessed state, the process of adhesion of the gold leaf traces to the substrate in the proposed technique may damage it and increase its resistance.Because the wafer paper used to fabricate traces is water-soluble, the adhesive layer applied to secure the traces gradually softens the paper.In the course of this adhesion, the wafer paper stretches slightly, molding itself to the shape of the substrate, whether curved or irregular.While this stretching strengthens the adhesion, it might cause damage to the gold leaf.For the conductivity evaluation, we therefore measured each sample's resistance, calculated the sheet resistance in Ω/sq, and then averaged these results for each condition.
In the following sections, we frst report the durability of the gold-leaf traces against tensile strength and surface roughness.We compared gold-leaf traces using wafer paper of three diferent thicknesses.If the traces exhibit durability against tensile strength, they are easier to handle during the adhesive process.Similarly, if the traces are durable on rough surfaces, they are considered adaptable to foods with uneven surfaces.We then report whether the goldleaf traces could exhibit adhesion and retain conductivity to foods with a variety of surface characteristics.We further evaluated the long-term adhesion and conductivity of the traces.We summarize our key fndings in this section to help predict and guide the design procedures for researchers and manufacturers looking to adopt our technique in their applications.

Tensile Strength
We prepared gold-leaf traces using wafer paper of three diferent thicknesses (0.01 mm, 0.03 mm, and 0.2 mm, as described in Sec.3.2.1).We used the fabrication technique to form simple linear traces with circular connectors at both ends.These traces were then adhered to 3D-printed test geometries.As shown in Fig. 5 (a), the geometry was designed to stretch the central sections (2 mm wide and 10 mm long) from 0% (no stretching) to 5% in increments of 1%.This method of evaluating tensile strength follows the approach outlined by Groeger et al. [22].We tested 3 (thickness) × 6 (tensile length) × 3 (samples) = 54 samples in total.
Results.We confrmed that traces made with wafer paper of all thicknesses successfully withstood stretching up to 4%. Figure 5 (c) shows the increase in resistance before and after stretching.Across all thickness conditions, we noted an increase in resistance with stretching.The magnifed view in Fig. 5 (b) reveals that the traces under the 0.01-mm-and 0.03-mm-thick conditions exhibit several fne cracks in a direction perpendicular to the direction of tension, whereas the traces under the 0.2-mm-thick condition have comparatively less damage.

Roughness of Substrate Surface
As with the above tests, we prepared traces using wafer paper of three diferent thicknesses (0.01 mm, 0.03 mm, and 0.2 mm).We adhered the traces onto sandpaper of seven diferent grit sizes (ISO/FEPA Grit designation from rougher to fner: 40, 60, 80, 100, 120, 180, 240).The grit size is a rating of the size of the abrasive Results.All samples showed good adhesion with no partial peeling of traces from the substrate.Figure 6 shows the results for the conductivity of traces under each grit size condition.While the change in resistance was negligible for the 0.2-mm-thick condition (maximum increase of 0.04 Ω/sq), resistance increased considerably with the increasing roughness under the 0.01-mm-and 0.03-mmthick conditions (maximum increase of 1.31 Ω/sq and 0.82 Ω/sq, respectively).We assume this is because the thicker the wafer paper, the better it flls in the irregularities of the sandpaper surface during the adhesion process, thereby reducing the impact of the unevenness on the gold-leaf layer.

Tests on Food Surfaces
We tested the adherence of the traces on various food surfaces to determine how their surface properties afect our technique.Since an exhaustive evaluation of all food materials is beyond the scope of this work, we selected nine common foods (see Table 2) that are familiar to most people, while also including edge cases with potentially challenging material properties.In addition, we ensured that the selected foods had variations in textual properties, elastic properties, and water interaction properties.In addition to surface roughness and elasticity, the water-related properties of the food are considered important.This is because the adhesive layer contains moisture, which can afect the adhesion of the trace.In addition, oil present on the surface of the food may interfere with the adhesion of the trace.
We adhered the traces with 0.01-mm-thick wafer paper on the foods, and when either or both adhesion or conductivity were problematic (i.e., peeled or electrically disconnected), additional tests were performed using traces with 0.03-mm-and 0.2-mm-thick wafer paper.The width of the traces was 0.5, 1, 2, and 3 mm.For the 0.01-mm-thick condition, we tested 9 (foods) × 4 (trace width) × 3 (samples) = 108 samples.Additionally, we tested 54 samples for the 0.03-mm and 0.2-mm conditions.
Results. Figure 7 shows the conductivity and adhesion results for traces on the foods.As shown in these images, all samples exhibited good adhesion.Regarding conductivity, under the 0.01mm-thick condition, traces of 0.5-mm width were broken in several foods.In foods with pronounced roughness, such as textured cookies and cereal bars, even 1.0-mm-wide traces were broken.Through additional testing for the 0.03-mm-and 0.2-mm-thickness conditions on cases where traces were broken in the 0.01-mm-thickness condition, we confrmed that conductivity could be maintained in all cases in the 0.2-mm-thickness condition and in all cases except the potato crisp in the 0.03-mm-thickness condition.
Results.In all combinations of conditions and thicknesses, we found that the traces did not change signifcantly in resistance (maximum increase of 0.19Ω/sq) and could be preserved under a variety of temperature and humidity conditions for the period typically used to preserve food.

Discussion
The results of our technical evaluations showed that the trace adhesion was consistently good.The proposed technique was adaptable to all of the water-interaction, textural, and elastic properties of the substrate with respect to adhesion.Regarding conductivity, durability to tensile and unevenness was higher for traces made with thicker wafer paper.The same trend can be confrmed from tests conducted on the adhesion of traces to various foods.On the other hand, excluding some edge cases (such as a potato crisp, which has a highly jagged or ridged surface), traces made with 0.01-mm-thick wafer paper could maintain their conductivity on most food surfaces.The measured resistance for the traces is still in an acceptable range for prototyping many types of interactive elements (such as those discussed in Sec.3.1.2).

PSYCHOPHYSICAL EXPERIMENT
Since the FoodSkin technique can potentially afect the original taste and texture of foods, it is necessary to evaluate how users perceive foods with gold-leaf trace applied.We investigated how the taste and texture of the food difered when gold leaf traces were adhered to the food compared with unprocessed foods, and the efect of the thickness of the wafer paper.

Participants and Experimental Setup
We recruited ten healthy participants (fve women, fve men, average age of 24.0 with SD = 3.0).The experiment was conducted following the safety standards approved by the local ethics research committee at Ochanomizu University.Participants were paid $10 in accordance with our organization's regulations.The experiment was explained to the participants prior to their participation, and they signed letters of consent.The experiments were conducted in a quiet room with a room temperature of 24-25 • C. We chose chocolate as the experimental material because it is widely consumed, making it easier for participants to evaluate their experiences with Table 2: Surface properties in tested food samples.A check mark indicates that the food possesses the specifed property.Asterisks indicate properties within the categories that are not mutually exclusive and may coexist with other properties.Note that these characteristics are specifc to the individual products we tested and should not be generalized to all products of the same type.

Water Interaction Properties Textural Properties Elastic Properties
Absorption Repellency Solubility* Texture Smoothness Oiliness* Softness Stifness Cookie (smooth)  taste and texture.In addition, since chocolate melts on the tongue due to body temperature, we considered it to be susceptible to the infuence of the presence of gold leaf and an adhesive layer between the tongue and the food.We prepared gold-leaf traces using wafer paper with three different thicknesses (0.01 mm, 0.03 mm, and 0.2 mm), similar to the technical evaluation, and adhered them to the underside of a piece of chocolate.In our preliminary study involving fve participants (none of whom participated in the following formal evaluation), we found that under the 0.2-mm-thick condition, the presence of the trace was distinctly noticeable, even without a comparison to unprocessed samples (with 100% accuracy by all participants).Thus, we decided to evaluate wafer paper with thicknesses of 0.01 mm and 0.03 mm in this experiment.Hereafter, the food processed by the proposed technique using wafer paper of each thickness is referred to as 0.01 and 0.03 , and the unprocessed foods as .

Procedure
Participants were seated in a chair, blindfolded, and ftted with nitrile gloves (Fig. 8 (a)).As shown in Fig. 9, this experiment consists of two parts (comparing (1) 0.01 and , and (2) 0.03 and , Fig. 8 (b)) and two sections (Taste and Texture).We conducted the two parts on separate days.Each section is divided into two phases (a memorization phase and a comparison phase).The sections were counterbalanced.
Memorization phase: First, depending on the respective section, participants were instructed to memorize the taste or texture of unprocessed chocolates ( ) and processed chocolates ( 0.01 or 0.03 ).We provided participants with one sample from each condition, informing them which was the unprocessed and processed sample.Participants were then instructed to place a sample of chocolate on their tongue for fve seconds, chew it twice, and then spit it out into a paper cup.Participants placed the chocolate with gold leaf in their mouths so that the gold leafdecorated surface was in contact with their tongues.After trying each piece of chocolate, participants were instructed to rinse their mouths with water and spit it into a paper cup to eliminate any residual food remaining in the oral cavity.
Comparison phase: In this phase, participants were given a total of eight pieces of chocolate (four samples each for the processed and unprocessed conditions, in a randomized order), one at a time.Similar to the memorization phase, they were instructed to place it on their tongue, chew it, and then spit it out into a paper cup.As with the memorization phase, participants were instructed to rinse their mouths with water after each trial.
After each trial, participants verbally indicated whether each chocolate they tasted was processed or unprocessed, answering a two-alternative forced-choice question.Depending on the respective section, we instructed participants to decide on the basis of taste or texture.After all the trials, we conducted a questionnaire.

Results and Discussion
In total, we collected 8 (samples) × 2 (parts) × 2 (sections) × 10 (participants) = 320 answers.The accuracy calculated from the participants' responses is shown in Table 3.If the accuracy approaches 50% (the chance rate), it indicates that diferentiating between 0.01 and , as well as between 0.03 and , was challenging.We conducted a Wilcoxon signedrank sum test for the questionnaire results and found a signifcant trend between Taste ( = 2.0, = 0.0707), but there are no significant diferences in Texture ( = 13.5, = 0.5224).This suggests that 0.01 tends to be less distinguishable in taste than 0.03 .Figure 10 shows the accuracy for each participant.Here, we can see individual diferences: several participants answered correctly   with high accuracy (P6, 9, and 10), while others had almost chance rates of correct answers (P2, 3, 4, and 5) for each condition.More than half of the participants commented in the post-questionnaire that there were trials in which they "were not confdent in distinguishing between processed and unprocessed" (P2-5, 8, and 9).The participants with high accuracy commented that they were able to discriminate conditions due to "subtle diferences in favor or texture occuring due to how the chocolate melts upon contact with the tongue" (P6).On the other hand, they also commented, "Without the prior learning (memorization) phase, I would not have been able to tell from the taste or texture" (P7).
In summary, the statistical analysis suggests a tendency toward 0.01 , ofering a taste that is somewhat closer to the original compared with 0.03 .Based on participants' feedback that "the trace was positioned right between the tongue and the chocolate, making it sometimes difcult to perceive the taste of the chocolate when it melted, " we consider this trend is likely related to the amount of wafer paper used (i.e., the 0.03-mm-thickness condition is three times the amount of the 0.01-mm-thickness condition).Considering that the chance rate is 50% accuracy, the average classifcation accuracy for taste and texture under each condition is modest (65.0-78.8%).In addition, participants' comments revealed that not a few of them were not very confdent in their answers; we assume that the diferences between the processed and the original were not that obvious.

APPLICATIONS
On the basis of the characteristics of the materials and the fabrication technique confrmed in the above evaluations, we present several demonstrative applications using dry food products as substrates.The circuit diagrams for each application are included in the Appendix.

Multipoint Electrical Stimulation
Figure 11 (a) shows the application of presenting multipoint electrical stimuli in the oral cavity.Our technique enables the placement of electrodes on areas of the food surface that directly touch the tongue.This allows for presenting electrical taste sensations at multiple arbitrary points simultaneously in foods that are tasted entirely or primarily in the oral cavity, such as candy.Although existing techniques (e.g., transcutaneous electrical nerve stimulation) can be used to present tastes at arbitrary locations [57,71], these are limited to a single location and cannot present tastes at multiple arbitrary locations simultaneously.Note that while prolonged licking leads to candy dissolving and changing shape, which can result in the peeling or damaging of the gold leaf, we have verifed that the gold-leaf circuit can still stimulate the oral cavity for a certain period (at least a few minutes) after licking starts.
In this application, we apply the direct electrical current to the human tongue via gold-leaf traces placed on the chocolate candy surface.Typical electrogustometers are designed to output direct current at 4 µA to 300 µA based on human detection thresholds [53].In our implementation, we use a power supply with a direct current of over 0.5 mA, which is sufcient to present electrical taste sensations to the user when the electrodes are in direct contact with the tongue.

Olfactory Enhancer
Figure 11 (c) demonstrates an application example of enhancing scent using a resistive heating element.This is done by utilizing the evaporation of volatile components present in the food.In this example, heating the surface of the almond cakes caused the aromatic compounds contained in the vanilla essence to evaporate more easily from the surface of the food, thus emphasizing the scent of vanilla.Note that the volatilization temperature of the components in vanilla essence is largely dependent on the volatilization temperature of ethanol, which is a major component.While the boiling point of ethanol is approximately 78 • C, volatilization commences prior to reaching this point, and the rate of ethanol volatilization escalates with increasing temperature [45,73].We confrmed that our gold-leaf resistive circuit (line width: 1.0 mm, total length: 138 mm) is capable of adjusting the temperature to approximately 45 • C when a 6.0-V input voltage is applied, and to around 70 • C at a 10.0-V input voltage.Thus, in our application, by adjusting the applied voltage, it is possible to enhance the aroma in accordance with the temperature.

Sound-Enhanced Eating Experiences
Figure 11 (d) shows interactive chocolate candy using a capacitive sensing element that emits a sound from an external speaker upon consumption.The sound changes when the microcontroller senses the current that varies in accordance with the contact between the traces and the tongue or inside the mouth.This design ofers users a dynamic auditory experience while eating.For example, the crunch sound of a cookie can be audibly amplifed, or a charactershaped cookie can produce a character-specifc sound when bitten.These interactions were frst introduced in previous works (e.g., EaTheremin [33] and WeScream [76]), and the use of FoodSkin expands the scope of these techniques to include dried foods.

Monitoring Eating Behaviors
Figure 11 (e) shows a nutrition bar integrated with a resistive sensing element.When the user bites into the bar, a portion of the trace is damaged or removed, resulting in an increase in the resistance used to detect a bite.The system measures the user's eating speed on the basis of changes in the resistance of the trace and provides alerts if the eating pace is considered too fast.Monitoring eating behaviors, such as the amount and duration of food intake, can help prevent and manage health conditions such as obesity and diabetes [10,64].Furthermore, incorporating auditory and visual feedback based on the amount or speed of food eaten is known to be benefcial in educating children about proper eating habits [34].

WORKSHOP
To assess the accessibility of the FoodSkin workfow, we conducted a workshop (Fig. 13 (c)).We recruited one graduate and seven undergraduate students from a local university (fve women, three men, average age 21.0 with SD = 1.12).Four of the participants had previous experience with electronics.None of the participants had prior experience with the FoodSkin workfow.They were paid $10/h in accordance with our organization's regulations.

Procedure
We conducted a workshop in a studio-like lab space with the following sessions.
Introduction (20 min).The workshop started with the introduction of FoodSkin, our study scope, and application examples, and then we demonstrated the fabrication process of FoodSkin and the design support system.Session 1: Prototyping for simple line traces (100 min).Participants frst practiced making gold-leaf traces on a sheet of A-4 paper using the fabrication technique (30 min).After the practice, they were asked to complete simple line traces (shown in Fig. 12 (a)) on the food surfaces.The participants used wafer paper with three diferent thicknesses (0.01 mm, 0.03 mm, and 0.2 mm), similar to the technical evaluation.In total, they created six traces on the food surfaces within the allotted time (3 thickness × 2 trials × 10 minutes, for a total of 60 min).For each participant, thickness conditions were assigned in random order.At the end of the session, we asked participants about the difculty of the technique (10 min).Session 2: Design and prototyping for electrical taste circuit (50 min).We designed the sample trace pattern (Fig. 12 (b)) before the workshop, and the participants were asked to complete the following steps.1. Measure the size of the food using a ruler and enter the dimension into the design support system. 2. Use the system to design the pattern by imitating the sample pattern for a unipolar-type electrical taste circuit.3. Fabricate gold-leaf traces on the food surfaces.In this session, the participants used the 0.01mm-thick wafer paper.We provided various materials, including foods for the substrate (cookies, chocolate, nutrition bars, pound cake, langdosha, etc.) and tools (laser cutter, DC power supply, multimeter, etc.) for the FoodSkin fabrication process and confrmation of conductivity.At the end of the workshop, participants responded to a questionnaire, including the System Usability Scale (SUS) [8], to summarize their experience with using FoodSkin.The task design in this workshop was based on previous work [13].

Results and Discussion
Table 4 shows the time participants spent working on completing one trace (system-based design and manual fabrication time) and the number of retries due to trace break.All participants completed the replication tasks within the allotted time in all sessions (Fig. 13 (a)).We observed that they had no trouble throughout the design processes with the design support system, while some of them encountered the following technical issues during the manual fabrication step.Session 1: All participants successfully completed the task of fabricating and adhering traces using 0.2-mm-thick wafer paper without failure.However, with the 0.01-mm-and 0.03-mm-thick wafer paper, the total number of retries by all participants was 3 and 2, respectively.The reasons for these failures were as follows.
One participant (P5) accidentally adhered gold leaf to the tweezers due to static electricity and then damaged it while trying to remove it.Three participants (P4, P5, and P6) broke traces while using tweezers to pick it up.Additionally, one participant (P6) used too much adhesive when attaching the trace, which led to its curling.The trace was then damaged during an attempt to fatten it.Session 2: Two participants (P7 and P8) took a second trial to complete the circuit.One participant (P7) adhered the gold leaf to one side of the food, then pulled and broke the gold leaf as it was adhered to the adjacent side (Fig. 13 (b)).The other participant (P8) broke a trace while using tweezers to pick it up.
Participant feedback: Participants commented on the difculty of handling the gold leaf: "Gold leaf is delicate and fragile, so it is difcult to process" (P1, P2, and P8), "It was difcult to work with the traces across the corners of the food" (P2 and P4), and "When adhering the traces on soft food (pound cake), it was difcult to adjust the force of the tweezers" (P7).We found that most participants could use our design support system efectively.The SUS score was 80.6 on average (SD = 9.66), which can be considered "good" [8].Participants were generally positive about the system, commenting that it was "easy to learn and use" (P1, P8), "the 3D view helped me imagine the fnished prototype" (P5, P7), and "the operation was enjoyable" (P1, P4).On the other hand, several participants commented that they were frustrated by the restrictions shown by the system (e.g., a warning message to prevent wiring from short circuiting), and were not satisfed with the fexibility of the design (P5, P6).In our observation, several participants checked the position and direction of the gold leaf by using the system's 3D view when applying it, and the system was also used efectively in the fabrication process.
In summary, we verifed that all participants learned the Food-Skin techniques and completed the tasks within the allotted time.From the results of the number of trials, the questionnaire, and the participants' comments, we conclude that users can adopt our design support system without any special training, but the manual fabrication process requires some profciency training (several hours).Although the thicker wafer paper required less time to complete the work and resulted in fewer failures, all participants were able to complete the circuit fabrication even when using 0.01-mmthick wafer paper.

LIMITATIONS AND DISCUSSION 8.1 Adaptability
A comprehensive investigation of the combination of types of foods, trace shapes, and attachment locations is beyond the scope of this paper.Also, the foods we chose were primarily carbohydrates, since vegetables and meats tend to be high in moisture.Although we tried to include food with a wide range of properties, one of the limitations of this study is that the results are sample-based.Most foods within our scope of investigation were feasible for applying our technique.One edge case we identifed was potato crisps, characterized by their distinctive shape, pronounced surface unevenness, and doubly curved surfaces.we confrmed that the traces worked better on smooth surfaces, even those with doubly curved geometries (Fig. 14 (a), (b)), suggesting that pronounced surface irregularities pose the most signifcant challenge to our technique.On the other hand, we confrmed that using thick wafer paper for tracing creation can prevent trace disconnections even with surface unevenness similar to that of potato crisp.

Durability
Our technical evaluation showed that FoodSkin can be applied to foods with varying textures and curvatures.We also confrmed that after drying and adhering, the gold-leaf traces are not easily damaged even with scratching or rubbing, not just a gentle touch.However, as shown in the mechanical evaluation in Sec.4.1, goldleaf traces lack durability during stretching.This could make it challenging to use them with foods like mufns or cookie dough, which alter shape when baked.Although we investigated the adaptability of FoodSkin to commercially available pre-processed foods, our future work will include an examination of the adaptability of FoodSkin in the process of creating handmade pastries or dishes.

Feasibility in Home and Restaurant Use
One of the limitations of our technique is that laser cutters are not yet common in homes and restaurants.However, ongoing technological improvements and decreasing costs may alter this situation [19,27].Over time, laser cutters have become more afordable [19] (e.g., the laser cutter we used costs about $660), and their use in food processing has broadened [11,21,63,80].No longer just specialized tools, they have the potential to become as versatile in the kitchen as the multicooker or food processor in the future.
In our technique, we opted to use a personal laser cutter due to its consistent capability to produce traces with line widths as fne as 0.5 mm and to create multiple traces simultaneously.While less costly options such as craft cutters or design knives exist, they lack the same precision.Our preliminary tests showed that these alternatives could handle traces with line widths of at least 1.0 mm when using a design knife, and 1.5 mm or more when using a craft cutter7 .

Safety in Materials, Chemical Reactions, and Electric Stimulation
When gold leaf is ingested, it is usually not absorbed signifcantly by the body and is excreted without being metabolized [23].Gold leaf is widely regarded as non-toxic, stable, and chemically inert, indicating that it does not react with most substances [25].Although it is unlikely that applying electricity to gold leaf on food surfaces will change the gold's chemical composition, the food may undergo a thermal change when the gold leaf heats up.In addition, gold typically has a positive temperature coefcient of resistance.This means that its resistance increases as the temperature rises.However, gold's resistance change with temperature is minimal and generally stable.In our applications with the heating elements, no signifcant change in resistance with temperature has been observed.Some of our applications apply a low-intensity current to the human body.According to the safety guidelines for transcranial direct current stimulation (tDCS), exposing the human body to a maximum current of 4 mA for one hour is not considered to cause any harm [5].Compared to this standard, the intensity and duration of the current utilized in our application are signifcantly lower and shorter.Even if an individual consumes a large quantity of food processed with our proposed technique in one sitting, the duration of the current application is only a few minutes at most.This current generates a very low voltage due to the low contact resistance between the gold leaf and skin, or between the gold leaf, saliva, and skin.Therefore, we consider it highly unlikely that our application would have any harmful efects on the human body.

Electrical Connections
In our applications, it is necessary to connect the gold-leaf traces to external elements such as power supplies and microcontrollers.However, directly connecting inedible components to food items can be impractical in terms of ease of use and may afect user acceptability.To address this, some of our applications utilize implementations that separate food from external inedible components.In Fig. 11 (c), the connection between the traces on the food and the external circuit is established by attaching gold-leaf electrodes (connectors) to the plate.Although this is a simple implementation, more practical devices that incorporate electrodes and external circuits into tableware have been proposed [51], and the use of such devices would be more acceptable to users in terms of cleanliness or ease of use.This implementation is desirable for applications where heating elements are used.This is because the electrodes on the food are disconnected from the power supply when the user eats it, preventing the user's mouth from being burned by excessive heating.Similarly, a technique can be utilized in which conductive patterns are formed on the inner surface of the food package instead of the plate and connected to electrodes on the food.We utilized the Leaf Circuits technique [69] to form a pattern of gold leaf on the surface of plain paper in the prototyping of the package (Fig. 14  (c)).These methods require a DC connection; however, it is also promising to use AC connections (e.g., capacitive sensing [1,26]) or non-contact communication methods such as chipless RFID [14].

Efects of Wafer Paper Thickness
We conducted technical evaluations, psychological experiments, and user evaluations (through a workshop) using three diferent thicknesses of wafer paper.The results show that there is a tradeof between the sensory impact (taste and texture) and physical properties (durability and processing ease) of wafer paper related to its thickness.Increased thickness enhances its durability and ease of processing, which are positive aspects.However, this same increased thickness negatively afects the taste and texture, making it more noticeable.In contrast, while thinner wafer paper ofers a less pronounced taste and texture, which is benefcial for sensory quality, it compromises on durability and processing ease.Although priorities such as processing ease and sensory impact on the consumer will difer among users, we suggest using thick wafer paper when dealing with highly uneven food surfaces or when creating traces with a narrow line width (approximately 0.5 mm width).

CONCLUSION
We proposed FoodSkin, a technique for creating circuits on food surfaces using edible gold leaf, and described several promising applications.Our technique enables the incorporation of dry food into an electronic circuit, which is difcult with previously proposed techniques.We believe FoodSkin enriches the design space for the computer-augmented eating experience by enabling the digital fabrication of electronics on versatile materials, surfaces, and shapes of foods.

A CIRCUIT DIAGRAMS AND PACKAGE LAYOUT
In this appendix, we present the circuit diagrams for each of the application examples shown in Sec.6 (Fig. 15), as well as the layouts of the circuits formed in the packages that use the Leaf Circuit method [69] introduced in Sec.8.5 (Fig. 16).

Figure 2 :
Figure 2: Overview of the design support system.(1) Users select 3D models of food and choose an interactive element to integrate.(2) The system provides a 3D preview along with a trace editor.(3) The trace area is adjusted to move along the surface of the 3D model.(4) Finally, the system generates fabrication fles, with ofsets for cutting lines set to achieve the appropriate trace width for the following fabrication process.

Figure 3 :
Figure 3: Interactive elements supported by the design support system.

Figure 5 :
Figure 5: (a) 3D-printed geometry to evaluate diferent percentages of stretch.(b) Enlarged image of traces.The traces under 0.01mm-and 0.03-mm-thick conditions exhibit several fne cracks.(c) Increase in resistance of gold-leaf traces at each percentage of stretching after adhesion.

Figure 6 :
Figure 6: Increase in resistance of gold-leaf traces at each grit size after adhesion.

Figure 7 :
Figure 7: Results for tested foods with varying geometry and properties.

Figure 8 :
Figure 8: (a) Experimental setup.Participant places a chocolate in the mouth so that the gold leaf-decorated surface is in contact with the tongue.(b) Samples of unprocessed ( ) and processed ( 0.01 ).A simple gold-leaf trace is adhered to the bottom of the chocolates.

Figure 9 :
Figure 9: Overview of the psychophysical experiment.It consists of two parts (comparing 0.01 and , and 0.03 and ), and two sections (Taste and Texture).This fgure shows an example of the execution sequence.

Figure 11 (
Figure 11 (b) shows an application where a resistive heating element is formed on the inner surface of a fondant au chocolat (a French chocolate cake).Although the fondant au chocolat has a gooey, molten chocolate center when freshly made, it solidifes over time.Applying an electric current to the trace heats it, keeping the chocolate inside melted.Chocolate fully melts at around 30 • C [7], and we confrmed that our gold-leaf resistive circuit (line width: 1.0 mm, total length: 138 mm, input voltage: 4.0 V) can maintain a temperature of around 35 • C.

Figure 11 :
Figure 11: Applications of FoodSkin.(a) Multipoint electrical stimulation for the human tongue using gold-leaf traces.(b) Fondant au chocolat with a resistive heater element.Gold-leaf trace is connected to the battery via the electrodes formed on the paper cup.(c) Olfactory enhancer with the scent of vanilla.(d) Interactive chocolate candy that emits a sound from the external speaker connected to the instrument device when consumed.(e) Monitoring users' eating behaviors with a nutrition bar.The resistance changes when the user bites the bar.

Figure 12 :
Figure 12: (a) Trace patterns used in Session 1.(b) Samples for replication task in Session 2.

Figure 13 :
Figure 13: Samples of (a) successful and (b) failed traces created by participants.(c) Workshop scenes.