The MyElectric app has long been needing some work so that it both provides more useful information and also reports the daily kwh data correctly for different timezones and when the app is used as an always on display and the day changes.

There is now a new version available that adds a real-time power graph as well as totals for the week, month, year and all time and the daily kwh bar graph as before.

To generate the kWh per day graph this version focuses on supporting feeds either created with the "power to kWh" processor or "Wh accumulator" processor rather than the "power to kWh/d" processor and average power feeds. Trying to support each one was leading to quite a lot of bugs in addition to more fundamental issues with time zones present with the power to Kwh/d daily feed approach.

The "power to kWh" and "Wh accumulator" processors are both fully supported on the emonPi and emonBase running the low-write image. There's more on how this accumulating watt hour or kWh approach works here: Development: Calculating Wh totals on the emontx

I now have the new MyElectric app running alongside the MySolarPV app in my home and I'm really quite pleased with the results. The totals at the bottom and average kWh per day for each period gives me most of what data to see at a glance how my electricity consumption is fairing and having this information available on a wall mounted tablet really makes all the difference compared with having to login to emoncms on my computer to see the data.

I would like to make it easier to switch in between the different apps, a side swipe would be really nice. I would also like to look into a way to make it possible to download apps from an app repository from within emoncms itself so that new apps could be added without them having to be in the app repository on github, there's quite a bit of work needed to get there thought.

The emonPi updates its internal linux time from NTP when connected to the internet. However if the emonPi is to be used on an offline network or for an application when accurate timestamp is essential then a hardware Real Time Clock (RTC) can easily be added to the Pi's GPIO. We have tested using a DS3231 based RTC module. This RTC module communicates with the emonPi via I2C, it can be easily connected as follows by soldering a four-pin header onto the emonPi aux GPIO pins:

emonPi with hardware RTC module added

DS3231 based RTC module

Continuing with the investigation into the embodied energy that it takes to make an energy monitor, I thought I would explore in a little more detail the embodied energy of the microcontroller which is often associated with the most energy intensive aspect of electronics manufacture.


As I mentioned before I wouldn’t put a large amount of confidence in the accuracy of the figures below, computer chips are incredibly complex things that take a large number of manufacturing processes to make and I have found it hard to find open detailed information on energy requirements of these. The calculations below are based on the datasets that I could find including the EU ecodesign dataset which I'm told is one of the higher quality freely available sources. But I could not find much detail on how the figures where derived to know what it does and does not include and the breakdown of energy use in its manufacture.

The idea here is more to get an initial set of numbers that could be improved upon in the future. My motivation for looking into this started by being inspired by howies and patagonia's efforts on footprint and supply chain analysis, partly by reading sustainable energy without the hot air that identifies industry as being a significant part of overall energy use and also after reading the article here about 'the monster footprint of digital technology'.

The following starts by looking at the available datasets and then investigates deriving figures for the embodied energy of a chip by calculation of the actual semiconductor volume contained which compares well for the small IC embodied energy value.

1) Datasets

EU Ecodesign methodology
The EU ecodesign methodology downloadable here http://ec.europa.eu/DocsRoom/documents/5308/attachments/1/translations/en/renditions/native provides embodied energy values for a variety of electronic component categories, it provides two embodied energy values for IC's (integrated circuits) these are:

Large IC                        8021.88 MJ/kg
Small IC                        1786.73 MJ/kg

MJ/kg: 1,000,000 Joules per kg

The question is what constitutes a small or large IC? Dimensions? processing power? and there is a large difference between both of these values!

Another paper that lists sources for embodied energy values including the EU ecodesign dataset is this paper on the embodied energy of an offgrid light: http://users.humboldt.edu/arne/Alstone_etal_Lumina-TR9-Embodied-Energy_Jan11.pdf. It lists two other useful values:

Semiconductor Grade Si             35000 MJ/kg, Taiariol et al. 2001
EPROM Chip (M27C1001, 0.36 W IC)   12.5 MJ/chip,  Taiariol et al. 2001

Jean Claude Wippler of JeeLabs did an interesting post a couple of years ago on what is inside an ATmega8 chip. The Silicon part was extracted by dissolving the outer casing with sulphuric and nitric acid. http://jeelabs.org/2013/06/09/whats-inside-that-chip

The size of the semiconductor part inside the ATmega8 is 2855 x 2795 um, less than 3x3mm:


Is it possible to estimate the embodied energy of a microcontroller by working out the embodied energy of the silicon part and the casing part separately? and how would the figure compare with the embodied energy values given for a small and large IC in the EU ecodesign dataset?

Thickness
To find the embodied energy of the silicon part we need to first work out the weight of that part. To do this we will need the density and thickness of the silicon wafer. The thickness of a silicon wafer can be anywhere between 275um and 926um.

Density of silicon: 2.3290 g.cm3 https://en.wikipedia.org/?title=Silicon

Weight
2855 x 2795 um x 275 um = 0.00219 cm3 x 2.329 g.m3 = 0.0051 g
2855 x 2795 um x 625 um = 0.00499 cm3 x 2.329 g.m3 = 0.0116 g
2855 x 2795 um x 926 um = 0.00739 cm3 x 2.329 g.m3 = 0.0172 g

The embodied energy of silicon grade Si is around 35000 MJ/kg, 35 MJ/g

0.0051 g x 35MJ/g = 0.178 MJ = 0.049 kWh
0.0116 g x 35 MJ/g = 0.406 MJ = 0.113 kWh
0.0172 g x 35 MJ/g = 0.602 MJ = 0.167 kWh

The weight of a SMT Atmega 328 is 0.14375g:

Large IC: 8021.88 MJ/kg = 1.153 MJ = 0.320 kWh
Small IC: 1786.73 MJ/kg = 0.257 MJ = 0.071 kWh

The EPROM Chip (M27C1001, 0.36 W IC) example is 12.5 MJ/chip = 3.47 kWh
Which is between 10 and 50x the large/small IC estimates and 20 to 70x the wafer only estimates.

The large IC is between 2x and 6x the wafer only and small IC is between 0.5x to 1.5x the wafer only estimate.

The weight of the silicon part of the chip for the ATmega8 should be between 3.5% to 12% of the total chip weight.

The rest of the weight is metal and plastic. Copper has an embodied energy of between 38MJ/kg and 142MJ/kg depending on manufacturing process and plastic is anywhere between 80-120MJ/kg. If we assume a copper/plastic mix of around 100MJ/kg as a rough estimate.

This would then contribute an additional 0.0132 MJ = 0.0037 kWh which would only increase the embodied energy of the chip by around 2-7%.

Our estimate therefore for the embodied energy of the small microcontroller is between 0.053 kWh and 0.171 kWh, mid range of 0.117 kWh for 625um wafer which compares well with the small IC estimate of 0.071 kWh (it is of the right magnitude). We would expect a certain amount of assembly energy for combining the plastic, copper and semiconductor parts which is not included when taking a constituent parts approach which will likely add a little more to the total.

Package type?


Calculating the embodied energy this way brings up an important question about the package type of the IC, I.e how much plastic and connector metal surrounds the semiconductor core. If the size of the semiconductor inside a through-hole and SMT version of a ATmega328 is the same then surely a total weight based measure is not a good guide for the embodied energy. Given that over 90% the embodied energy is associated with 3.5 - 12% of the weight.

RaspberryPi chips
The ATmega8 or ATmega328 is not a particularly powerful chip, the emonpi also uses a RaspberryPi which has three chips on board.


Its not clear what the volumes of the semiconductors inside these chips are. But given their small size and relatively large processing power lets assume for an initial calculation that the size of the package is very close to the size of the semiconductor. If the wafer thickness is 625um then the embodied energy of the chips can be estimated as:

Broadcom        14 x 14 mm x 625um = ~2.8 kWh
Elpida          12 x 12 mm x 625um = ~2.0 kWh
Smst            8.7 x 8.7 mm x 625um = ~1.1 kWh

These estimates which cover a quad core 900Mhz + 1GB ram computer are much larger than the ATmega8 estimate as we would expect but are still significantly lower than the embodied energy figure given for the 32MB 2g memory chip referenced in the article on the the monster footprint of digital technology of 20 kWh - while delivering much more functionality, almost a full computer!

Are these figures accurate? are the underlying datasets accurate for modern processors? Does it mean that newer miniaturised technology with the ability to pack much more computing power inside the same volume of semiconductor has resulted in a significant reduction in the embodied energy for equivalent amounts of computing digital technology functionality? These are all significant guesses at the moment and it would be really interesting and useful to see more open data available on the embodied energy of newer technology. It would be good to see greater interest from the manufacturers of these chips in calculating and giving information on the embodied energy of their products, making data available on this would really help with understanding this question about the impact of technology.

Three weeks ago, time runs fast! I attended the open source circular economy event OSCEdays in London. It was a fantastic event with a lot of energy and ideas exploring how the circular economy could benefit from an open source approach.

OSCEdays at FabLab London

Initial draft embodied energy analysis of the emonpi

There was a wide range of challenges and projects there including: circular makerspaces submitted by the great recovery looking at how fablab's and makerspaces could encourage circular thinking, an initiative developing re-usable containers for cosmetics, a challenge looking at the implications of the circular economy for wearable technologies, 'trust is not waste' by the rubbish diet project and ours looking at the embodied energy and lifecycle analysis of the OpenEnergyMonitor EmonPi.

Id like to thank Rachel Stanley, David Green, WoonTan and Paidi Creed for all their help and input and for Erica Purvis and Sharon Prendeville for organising the event and the wider team that where involved in running the event in many other cities around the world.

Our discussions and findings are partly documented here, there's also a bit more that needs writing up which will be online soon: http://community.oscedays.org/t/headline-challenge-open-energy-life-cycles/662/2.

The research that I did on the embodied energy of the emonpi in the lead up to the event can be found here: http://openenergymonitor.blogspot.co.uk/2015/06/investigating-embodied-energy-of-emonpi.html

We are really interested in the lifecycle impacts and embodied energy side of OpenEnergyMonitor and so were going to keep building on this. I've almost finished a post that looks at the embodied energy calculation for the emonpi microcontroller and raspberrypi chips in more detail which will be up soon.

If your interested in embodied energy, lifecycle impacts, the circular economy and how open source could play a part do check out OSCEdays https://oscedays.org

The last 6 zero carbon energy model examples (linked below) have explored the core parts of a household energy model looking at supply and demand including the electrification of space heating demand and transport demand so that it can be provided for with renewable electricity.

This final model in this series brings all these components together to see how the combination of demands interact and how they affect the supply and demand matching.

It also explores the contribution of two small scale stores a 7kWh electrical store (such as a Tesla power wall) and a 10kWh heatstore.



The following table show the results in terms of percentage of demand supplied directly of running the model with different electric vehicle charging profiles, and storage options. There is a small 4% oversupply.

No stores
4% OS + night charging	62.3%
4% OS + 1/2day 1night charging	76.2%
4% OS + flat charging	78.3%
4% OS + smartcharging	86.3%
Liion store
4% OS + night charging + 7 kWh li-ion store	84.4%
4% OS + 1/2day 1night charging + 7 kWh li-ion store	86.0%
4% OS + flat charging + 7 kWh li-ion store	86.2%
4% OS + smart charging + 7 kWh li-ion store	88.6%
Heatstore
4% OS + night charging + 10 kWh heatstore	70.2%
4% OS + 1/2day 1night charging + 10 kWh heatstore	81.0%
4% OS + flat charging + 10 kWh heatstore	82.4%
4% OS + smartcharging + 10 kWh heatstore	87.5%
Liion + heatstore
4% OS + 7 kWh li-ion store + 10 kWh heatstore + nightcharging	85.5%
4% OS + 7 kWh li-ion store + 10 kWh heatstore + 1/2d 1n	86.9%
4% OS + 7 kWh li-ion store + 10 kWh heatstore + flatcharging	87.1%
4% OS + 7 kWh li-ion store + 10 kWh heatstore + smartcharging	89.3%

Its interesting that a matching of 89.3% can be achieved with 7 kWh li-ion store, 10 kWh heatstore and electric car smartcharging up from a minimum of 62.3% with no stores and night time charging only. I think its quite impressive and encouraging that this high a level of matching can be achieved from a relatively small amount of storage and that 86% can be achieved with the smartcharging option only.

There are clearly different routes possible to achieve higher degree's of matching. If smart charging is technically possible with the flexibility used within the model and that it doesn’t provide too much of a burden on the user and only requires potentially a relatively small amount of electronics, embodied energy and cost compared to the li-ion store and heatstore then smart charging may be a more effective option.

A li-ion store and flat rate charging provides about the same benefit as smart charging and so if smart charging does not pan out to be practical then there may be an option to make up for it with a li-ion store, especially if the embodied energy and cost of storage reduces significantly.

The combination of measures provide smaller gains (if you apply smart charging first then the li-ion store only provides ~3% additional gains, however if you apply the li-ion first it looks like its the smart charging that only provides the small gain). Perhaps an important aspect is that a combination could provide important redundancy. It looks worthwhile to explore and develop each of the above solutions with a focus on how they integrate, the flexibility at which they can match supply, their costs and embodied energy.

The other blog posts in this series are linked below and the next model will explore the use of large capacity energy stores such as power to gas to reach 100% supply/demand matching. All the modelling behind this work is open source and available on github here: https://github.com/TrystanLea/zcem 

• Modelling hourly demand and supply for renewable powered domestic electricity, heating with heatpumps and electric vehicles 
• Hourly energy model example 1: Variable Supply 
• Hourly energy model example 2: Variable supply and flat demand (python code included) 
• Hourly energy model example 3: Variable supply and traditional electricity demand 
• Hourly energy model example 4: Complementarity of different renewable generating technologies 
• Hourly energy model example 5: Simple space heating model with heatpump's powered by renewable energy 
• Hourly energy model example 6: Electric vehicles and a renewable energy supply

Continuing the blog series on building a hourly zero carbon energy model based on the ZeroCarbonBritain dataset the 6th example model explores another another key solution used in the ZeroCarbonBritain report and in David MacKay's book Sustainable energy without the hot air: the electrification of transport.

The intention here is to explore what level of supply/demand matching between a renewable energy supply and electric vehicle charging could be achievable with different electric vehicle charging profiles. A higher level of supply/demand matching reduces the amount of backup or energy storage required to meet demand.

The first example starts by integrating electric vehicles with a simple night time charge profile. The second example then explores more constant charge profile throughout the day – this constant charge profile could be the result of a large number of electric cars all charging at different times, some at work, others at home over night. The third example explores a basic smart charging approach where the charge rate can aligned with the availability of renewable energy. There are many people who are already choosing their charge times to align with domestic solar pv output and there is much discussion about the opportunity that this may provide for matching supply and demand.

Night time charging

The first model investigates night time charging between 1am and 8am (7 hours). The charge profile of an aggregation of electric vehicles is much more likely to be smoother than this with a distribution over the day especially as the number of work based charging facilities and rapid chargers increase but for interest we will consider this case.

	Onshore wind	Offshore wind	Wave	Tidal	Solar
Installed capacity	0.8699	0.5864	0.9931	1.1783	2.9825
Percentage of demand supplied directly	27%	28%	29%	28%	3%

Flat charging profile

The results of running the model for a flat demand profile is the same as the earlier example where we considered the degree of matching between supply and a flat demand. Substantial improvements in matching compared to the night time charging profile is gained by just managing to distribute the charging requirements evenly across a day. This is unlikely to be possible on a single household basis but perhaps possible when the aggregate demand of many hundred cars are taken into account with day time charging at work encouraged.

	Onshore wind	Offshore wind	Wave	Tidal	Solar
Installed capacity	0.8699	0.5864	0.9931	1.1783	2.9825
Percentage of demand supplied directly	65%	76%	74%	57%	40%
Percentage of time demand is more or the same as the supply	40%	46%	45%	38%	32%

Smart charging (variable rate charger that matches available supply)

Smart charging could allow electric vehicles that are left connected to the grid to charge when renewable electricity is available. The simple smart charging algorithm in this example starts by using available supply to charge the car's battery directly. A minimum SOC level required to cover the days journeys is maintained with a top-up charge if needed. The battery SOC is kept between 10% and 80% to help ensure long life is maintained.

The charge rate is based on a forecast of available supply over the next 24 hours, if the available supply is more than the forecast demand then the charge rate can be reduced. If there is twice as much supply forecast than demand then the rate of charge could by dropped to half the available supply in order to distribute the charge better across the 24h.

	Onshore wind	Offshore wind	Wave	Tidal	Solar
Installed capacity	0.8699	0.5864	0.9931	1.1783	2.9825
Percentage of demand supplied directly	78.7%	85.3%	80.3%	84.1%	72.2%
Percentage of time demand is more or the same as the supply	69.5%	74.5%	67.5%	76.1%	69.2%

The results show substantial improvements again for the addition of smart charging, with smart charging + solar showing the largest gain. It is notable that the model suggests that a 2.98 kW solar PV array (a fairly typical amount for a home solar install) could provide 72.2% of almost 10000 miles a year of driving directly.

The feasibility of implementing this kind of variable rate smart charging on a household level with onsite solar pv needs a bit more investigation. The domestic charger on the nissan leaf can vary its charge rate in between 7A and 32A for the 6kW model or 7A and 13A for the 3kW model. Several people have already build open hardware variable rate electric car chargers making use of the fact that its possible to send a low voltage signal to an electric car to request a charge rate. Here are a couple of links to electric car charging related discussions and resources:

Dod Davies solar charge controller
https://twitter.com/dodavies/status/541349518693117953

OPenEnergyMonitor based electric vehicle charging
http://openenergymonitor.org/emon/node/10805

Smart Charging a EVSE with OpenEnergyMonitor RF data, Working! http://openenergymonitor.org/emon/node/4930

Open EVSE:
https://code.google.com/p/open-evse/

From: https://twitter.com/dodavies/status/541349518693117953

There are many aspects to consider and understand better when building a smart charger for electric vehicles for example it is suggested that to prolong the health of the battery its better to charge at a higher rate and up to the moment of starting your journey (rather than charge and let the battery sit at a high SOC). A more in depth understanding of the consequences of implementing this kind of charging and the balance point between battery life impacts and improved grid stability benefits or improved household economics would need to be understood.

Another possibility is that an aggregate demand profile for charging hundreds of electric cars could generally fit a renewable energy availability profile through a mixture of a larger portion of fixed rate charging cars charging at times of high availability than at other times.

In the next and final example in this zero carbon energy modelling series, we will combine the demand models for traditional electricity demand, electric heating and electric vehicles into one model and explore the implication of different electric vehicle demand profiles when interacting with multiple generation sources and multiple demand types.

With my Refurbished Samsung Galexy Tab 3 energy display up and running I've made a few changes to the solarPV application specific dashboard to make it work better. Thanks to Steve for many of the suggestions, see forum post.

• The view now automatically updates as a rolling window, in any of the modes, 3 hour, 6 hour, day etc 
• The balance Import/Export is now the same size as the solar & house consumption. 
• Night time noise on the solar pv channel less than 10W is zeroed. 
• The in-window stats are now easier to see at a distance with larger font.
• The view buttons are easier to click on a touch screen.
• Its possible to make up the consumption or solar generation feed from multiple feeds on the fly by entering comma separated feed id's in the configuration interface.

Here's the result:


Upgrading
To upgrade on the emonPi or latest emonbase image navigate to the admin tab in emoncms and then click the update button. Otherwise the app's module can be downloaded and updated using git: http://github.com/emoncms/app

We've been discussing for a little while using re-purposed tablets for energy display's rather than trying to develop our own pre-assembled version of the EmonGLCD (we're planning on keeping the through-hole version though).

I got myself a refurbished Samsung Galaxy tab 3 last week for £50 off ebay and the Koala Tablet Wall Mount Dock by Dockem.

Here are a couple of pictures of it in action:

It's fantastic when we get top hear about our energy monitors being used for applications we have never have thought of. Here is one such example:

SolarMill Writes:

We've just published our first open source project! It's called OpenEMC, and it's a code modification for the emonTH sensor by the OpenEnergyMonitor project. OpenEMC translates temperature and humidity readings into an easy to understand equilibrium moisture content (EMC) value, useful for woodworkers and operators of solar kilns.

We’ve been using Open Energy Monitoring components for the past few months for power monitoring and love its open source flexibility.  We recently received an emonTH to monitor Temp and Humidity values in our workshop and have developed a useful modification to the firmware.

emonTH code is on GitHub: https://github.com/solarmill/OpenEMC/tree/origin/emc

Read more about this application on this technical and well written forum post by Bert Green and Andy Fabian.

emonTH in drying box

Stable EMCin Controlled Box

SolarMill make eco-friendly gifts and home decor using solar-powered machinery, they have a super cool looking workshop:




http://www.solarmill.com/

The 5th example in the online zero carbon energy modelling tool is where it starts to get more interesting: modelling the supply/demand matching between a variable renewable supply and space heating delivered by heatpumps. The electrification of heating with heatpumps so that heating can be supplied with renewable electricity is one of the key solutions used in ZeroCarbonBritain and Sustainable energy without the hot air.


Online tool: http://openenergymonitor.org/energymodel > navigate to 5. Variable supply and space heating demand

The ZeroCarbonBritain dataset includes weighted daily temperatures for 10 years. This dataset can be combined with the solar dataset and a basic household energy model to create a seasonal heating demand model.

Solar gains are an important aspect of space heating. Using MyHomeEnergyPlanner (The open source retrofit modelling tool we are developing with Carbon Coop) to model a concept low energy house with fabric energy efficiency of 120W/K and a total of 16m2 of window area on a 80m2 (floor area) house with external surface area 206.4m2 the maximum potential annual solar gains where calculated to be 4429 kWh.

MyHomeEnergyPlanner: http://github.com/emoncms/MyHomeEnergyPlanner

Running a low electricity demand of 2200 kWh a year and taking into account the solar gains. The space heating demand is only 4,247 kWh/year (compared to 13500 kWh/year for a typical house today a 68% space heating energy saving). If that remaining heat demand is supplied by a heatpump the electrical input should be 1,416 kWh/year.

Building an hourly space heating model:

In order to calculate the space heating demand the model first calculates the total heating demand before solar gains and internal gains are taken into account. The space heating demand assumes constant internal temperature target of 18.5C rather than a morning and evening heating period. A further example with a higher internal temperature and variable profile would worth exploring for comparison. (21.0C being the passivhaus internal temperature target and there is an interesting discussion about the role of heating profiles, heatpump performance and demand spikes)

The model then subtracts the internal gains (heat given off by appliances/cooking/lights etc) and the heat provided by solar gains through the windows, we use the solar pv capacity factor dataset here to provide our solar irradiance dataset. The amount of solar gain was scaled to match the amount of solar gain calculated in the SAP MyHomeEnergyPlanner tool based on the window orientations and areas – the total available solar gain energy is equivalent of 5.0 kW of solar pv.

The model also keeps track of unused solar and internal gains when the internal temperature is already at the target temperature. The assumption at this point is that the excess heat is dumped outside perhaps through increased ventilation.

The space heating demand after internal and solar gains are taken into account is then supplied with a heatpump with the simplifying assumption that the COP is constant and the heatpump fully responsive. A more complex model taking into account a degree of thermal mass in the building and the dynamics of the heatpump cross checked with real data would be useful here to check the assumptions taken in order to create an initial simplified model.

Running the same fabric energy efficiency, max available solar gains and internal gains through this hourly model gives a space heating demand that is 14% higher than the space heating demand calculated with the SAP based MyHomeEnergyPlanner. The difference may be due to the differences in the way solar gains and internal gains utilisation is calculated in a monthly vs hourly model, further investigation is needed to fully understand the reason for this difference.

Model heating demand results:

Total heat demand	8445 kWh/y
- Total utilized internal gains:	2044 kWh/y of 2201 kWh/y
- Total utilized solar gains:	1566 kWh/y of 4132 kWh/y
= Total space heating demand:	4835 kWh/y
Total heatpump electricity demand:	1611 kWh/y

Running the model for each renewable generation type to investigate the degree of direct supply demand matching we get the following results:

	Onshore wind	Offshore wind	Wave	Tidal	Solar
Installed capacity	0.568kW	0.383kW	0.651kW	0.776kW	1.95kW
Percentage of demand supplied directly	51%	57%	61%	43%	10%
Percentage of time demand is more or the same as the supply	59%	59%	58%	54%	41%

Onshore, Offshore and Wave give quite similar levels of matching. Solar PV supplies the least demand because most of the solar electricity is generated in the summer and most of the heating demand is of course in the winter but also importantly when the sun is shining the heat demand is less due to direct solar gains, the dataset we are using for solar pv generation and solar gains is the same dataset.

The online example also explores the effect of adding a very basic thermal store in order to increase the level of supply matching.

The source code and datasets for the heating demand model and full supply/heating demand matching simulation is all open source available in both javascript and python.

Space heating demand varsupply_spaceheatingdemand.py
Space heating demand with heatstorevarsupply_spaceheatingdemand_store.py
Full source code: https://github.com/TrystanLea/zcem