The next Tetrinsic concept was going to be "Tetrinsic Concept4", but partially because this project needed a hero and partially because I was thinking about the intro music from the 33 second mark of the Fantastic Four: Worlds Greatest Heroes as I was thinking up all the improvements I could make from Concept 3.2X2, the internal code name is "Tetrinsic Concept Fantastic4".

The idea that forms the basis of this concept was from @RunnerPack all the way back in January, and is something I didn't really want to work on until after I actually had a proof-of-concept to test since it would involve more 3D printed / custom components. I wanted to be able to compare the off-the-shelf hardware solution to the more custom made solution. 

Now that the 580KV motor is no longer available and I know that cogging should really be 0, I'm going to have to go through with a custom motor solution. While I was at it, I wanted to see if I could increase the stroke : body length ratio.

Initial thoughts

I thought about a solution for about an hour, and this is what I envisioned:

• Tetrinsic becomes a linear BLDC motor with integrated hall effect and pressure sensors.
  • Part of the reason is so that I have more space to make and place coils.
• I move all the logic back to #Tetent [gd0090] / #Tetrescent [gd0150] like I originally planned at the start of Tetrinsic Concept1
• The motor controller is already designed to let me bundle 4 motors to one CS pin, so I just need 2 CS pins for all 5 linear motors
• I can use the 3 hall sensors that are sometimes installed into BLDC motors as the position encoder.
• I only need 2 bent tubes, and the turning pulley can fit inside the loop instead of outside
  • This assumes I can create a surface that can turn on a 7mm diameter radius
• Coils will be on the top and bottom of the linear motor section, which makes the magnetic field much more consistent and means that I don't have to worry too much about air gap
• The belt can be printed as a "Spiralise Outer Contour" (vase mode) of TPU
  • The guy on the Mass Production Hack Chat said that TPU was the best in terms of layer adhesion.
  • It can also be made from a magnet-band-magnet sandwich, where 2 separate magnets are stuck on each side of a thin loop of material.
  • I later realised that this would mean that the Thumb Tetrinsic would have to be a different height than the Finger Tetrinsics so that the loop could go around them.
• On the connector board, I can use a 5 input Sigma-Delta, 24bit ADC for the load cells and up to 5x multi-channel 16 bit adc's for the hall effect and current sensing 
  • (I've been hearing multiple times that the ADC on the esp32 isn't great)
• Encasing the top-down coils in a ferrous material should further increase the field strength between the coils and the magnet embedded belt.
• See this video for more information.

My idea is to take a linear motor like the one shown below, and swap the coils and electromagnets:

The design shown above also looks like a scaled up version of a voice coil actuator, which can be used for haptic vibrations.

Anyway, cogging is eliminated because there's nothing magnetic in the stator. The first time I saw the video about cogging, I didn't fully grasp it, but the thumbnail actually does a good job of showing what's actually going on:

From the perspective of the magnets, the copper coil may as well not exist. The ferrous stator teeth it's wrapped around is a different story. Because of the gaps between the stator caps, there's actually an imbalance of attractive forces. (remember, magnetic materials are attracted to both north and south poles of magnets). I'd imagine that, to solve it, you'd either need to remove the ferrous stator entirely or design the stator such that the attractive force is equal across the entire perimeter.

Magnets

The first reason why Me In The Past didn't think that the embedded magnet idea would work was because I was mentally simulating with 1.5mm cube magnets. This was back when I was working on Tetrinsic (Concept) 2.0.

I measured the magnet length in the 580KV rotor to be 8mm and the stator laminations to be 7mm. Thus I started looking for an 8*2*2mm magnet. That doesn't really exist, but an 8*3*2mm, 10*3*2mm and 10*3*3mm magnet does. Here are some that I found:

Then I sketched up an rough estimate for the overall loop:

313/4.2 = 74.5238. I require "pole pairs", which means the amount of magnets needs to be multiples of 2. Thus, the min amount of magnets needed for the belt is 76.

I went with 10*3*2mm magnets because the motor strength should be 25% higher, I've got a better chance of fitting the pulley in-between the sliding rods and I have the opportunity of going to a 10*3*3mm magnet if needed. For now, the 2mm magnet as the benefits of being able to obtain a slightly shorter Tetrinsic and it's more obvious than the 3*3 which faces have poles and which are just the sides. The belt would also be lighter, which means external forces would have less of an effect on the load cell readings. More on that further into this log.

This is what the belt would look like on the steel rods:

Print Material

The obvious choice, at least on paper, is PBT:

Easy printing? Doesn't even need a heated bed yet has a working temperature of over 130 degrees Celsius? High inter-layer bonding and toughness? I haven't tried it, but it's essentially obtainable unobtainium. I even tried making my own. I asked the seller what the grey looks like and I believe it looks great:

From my experience trying to extrude pellets into filament, I can only imagine the only imperitive step before printing this material is to print a temperature tower. This material has a very sharp melting point.

Accelerometer

The mass of the assembly responsible for measuring the pressure of each finger is now higher, meaning it's even more likely that acceleration and orientation changes will cause Tetrinsic to think that it's being pressed when that's not the case. Thus, I need a 1 axis accelerometer that's in the same axis as the force sensing part of the load cell. 

Long story short, the BMI160 is still the best chip for this. I looked into the sample rates, and it looks like I should target 800 or 1600 samples per second:

The amount of samples per second directly affects the latency of the overall system. The last thing I want to find is that my fingers are feeling a haptic event for one layer but the force I'm applying means I'm already in a layer above / beneath it.

As long as the data is clean, 800SPS should be fine.

New Feature: UVC sanitation

Whilst researching and planning out this concept, I had a realisation. The belt can be electronically moved out of view, I could place a UV LED on the underside and heavily touched items such as keyboards and smartphones gather a lot of microorganisms. I also remember the below video I recently watched on "Big Hand Dryer" vs "Big Paper Towel", namely the part when Half As Interesting said "[Big Paper Towel] hit 'em on hygiene.".

I quickly searched to see if UV LEDs have been used to sterilize keyboards, and I found this video from Kensington on a product called the UVStand:

Then some quick research on wavelengths, protective eyewear and purchase options followed:

Coil Creation

I knew that the coils were going to be the most tedious out of this entire operation, so I wanted to look into strategies. 

This video on YouTube said that coils of copper inside the perimeter of the magnet wouldn't generate any energy (he's making a wind turbine), and seems to be confirmed by looking at a StackExchange question:

For this application, the Kv should be as low as possible. This means that the turn count of the motor needs to be as high as possible. However, the higher the turn count, the longer it would take to coil, maintaining coil shape quality.

On my research travels, I found this image:

The coils were actually bought:

It doesn't seem like they'd have the size I'm looking for:

So I did some more mental simulations and for conductivity of heat, electrical resistance, number of turns, ease of coil manufacture and quality of coil geometry, I'm not really seeing the drawbacks to using copper foil here:

I looked around and found one with a datasheet, called the AT0525 35 Micron Copper Foil Shielding Tape. It's got a foil thickness of 0.035mm and an adhesive thickness of 0.03mm for a total thickness of 0.065mm. I did some more research and found out about orthocyclic wound coils.

The amount of spacing needs to be 3 coils for every 4 magnets.

I'm planning on using a 4.5mm magnet-to-magnet spacing, which means that the max coil width can be 4.5mm*4/3 = 6mm.

I found this 3mm foil that is only conductive on one side which has a thickness of 0.05mm and a length of 30 metres. I also found this table that gives me a rough idea as to the turn count relationship to Kv:

It seems that I can fit about 28 turns and still have space for an insulator such as Kapton tape to go around the entire coil.

The middle set of 3 coils are removed, and in their place is the 17.5*17.5mm PCB(s) that have the accelerometer and UVC LED. As you can see, each Tetrinsic would use 36 coils, meaning that I'd have to make over 360 of them for 2 Tetrescents, using over 305 metres of foil in the process. Not having to precisely wind thousands of metres of 0.1 - 0.2mm wire onto a 3mm tall coil is the main reason of attempting this foil method. It should be possible to create 9 foil coils at once if I design a suitable jig. 

The Analog To Digital Converter

This was the part that took the most time and research. This was mainly because it was the difference between Tetrinsic costing £17.20/ea and as much as £23/ea (which is a difference of £50 for 10 of them). 

The first thing I did was search Digikey and found out that the ADCs that I can use still start at over £3/ea. Thus I quickly turned to AliExpress and found the ADS1256:

£2 is kind of odd, since on Ti's site, it's like $12:

Well there were many sellers selling it at under £2/ea, with the cheapest being £1.50 after tax:

The first things I noticed was that I could save a clock pin and use a crystal (though I'd rather avoid it) and that there was even an IO extender in here:

The ADS131 series is a lot rarer on AliExpress. I couldn't find the M06 but I was able to find the M08 edition:

This is £5.80/ea after VAT.

I then found the ADS1262, which is a 32 bit ADC

This thing also has IO, but it's shared with the ADC inputs. 

Now, the ADS131 series is the only ADC I've found that can simultaneously sample. The other two, as well as the MCP3564R I'll mention shortly, only have a single ADC. There's also the ADS1263 that has a 32 bit and a 24 bit ADC, but the 24BIT ADC can only do 800SPS (which I was planning to split across 3 current sense pins), the datasheet suggest that this ADC has less than 12 bits of effective resolution at that full speed, and costs about the same as the M08 on AliExpress. This is one notable slide from Ti that I saw when considering my options:

Moving on, the less things I need to sample, the better. It turns out that I could only sample 2 of the 3 sense pins and then calculate the current in the 3rd one:

I'd imagine that I can also have 2 hall effect sensors for position:

A little side note is that these sensors have more in them than I thought:

There's the below linear motor that, even though it's only got a 34mm mover length and 2mm thick magnets, it can generate up to 10N of continuous force and also uses hall effect sensors to obtain position from 2 signals:

Back to the ADCs, that essentially meant that I have 5 or 6 sensors to scan, meaning that I need 4000 - 4800 SPS if aiming for 800 SPS across all channels. Below are the Effective Number Of Bits (ENOB) of some of the options.

32 Bit Multiplexed:

24 Bit Multiplexed:

24 Bit Simultaneously Sampled:

And some research suggested that to convert Effective Resolution to Noise-Free Resolution, I take away 2.7 from the number. Looking at the difference of the 24-bit mux, I believe that is the case.

I then looked into scanning the inputs, and found this in the 32 bit datasheet:

1/4800 * 1000 = 0.2083 ms. The only data rate option that can get that is the 38400 SPS option. Looking at the noise free bits, it tanks to 12.5 across the board doing that.

The 19200 SPS option has a settling time of 0.337ms, resulting in a top frequency of 2967Hz. I'd have to go down to 400SPS to use this. That still exceeds the touch sampling rate of most gaming smartphones (360SPS), but I'm not sure if it's fast enough for haptics. It also increases the latency between the measurement of the sin and cos hall effect sensors. The above table uses an external 5V reference, and it sounds like the reference voltage only affects 38400SPS:

For the 24 bit mux, it's maximum is 4374Hz:

By sampling 2 sense pins, I could still obtain 800SPS and I'd get 15 bits at PGA = 32.

The reason why I'm looking at the noise-free bits is because of this:

Now, the load cell (well, a very similar load cell to the one in the precision scales) states that it's excitation is 1mV +/- 10%. That seems to mean that with a 3.3V source, I'd get +/- 3.3mV of full scale signal. The below is the 32 bit ADC's mention of PGAs:

That's like... 78mV. For the 24 bit mux, it's the same for the highest PGA value:

The issue is that it actually doesn't come with a reference voltage source:

I can't use 3.3V for the positive and GND for the negative terminal because Vref has to be less than 2.6V. Also, the buffer needs to be kept off unless I want to be using something like a 1.25V reference voltage.

Overall, there's something like 12 components I need to also allocate PCB space for to even use this chip. I could always put this on the connector PCB, but I'm not sure if it's a good idea having a reference voltage trace spanning across the entire device.

Moving on, it turns out that the M08 and M06 share the same packages and pins, just that the latter (on the right) has 4 more NC pins.

I'm not sure why Ti made seperate datasheets for all the chips in the ADS131M0x range. It's also a bit painful that they said it's "optimized for cost sensitive applications" when it's easily the most expensive chip:

Anyway, most everything in them are identical and there's something I noticed about the ENOB:

As you can see, PGA=32, 64 and 128 don't exactly do anything. The signal gets doubled, but then you lose 1 bit of precision. It's essentially like just multiplying the values by 2 in software. The FSR at 128 is +/- 9.375mV.

Side tangent: There's also chips that sense voltage and current:

I thought it was quite cool.

Back to ADCs, I found the reason why I wasn't finding the MCP3564R:

Even though Digikey says it has 4 inputs, it's actually "4, 8" inputs. I've reported it.

[11 Sep: I just got an email back saying they've fixed the whole series and should be reflected on the site shortly.]

Anyway, the datasheet seems to imply that it's specifically designed to be able to multiplex through inputs without settling times. The ENOB values for 32 and 64 have also been omitted due to the 1-entire-bit reduction I mentioned.

This chip has enough speed to sample 6 channels at 1600SPS whilst still allowing for a full 24 bits of resolution:

I think it would make sense to get 5 inputs at 1600SPS and then scan other inputs for diagnostics information, such as the solar cell voltage or ADC temperature.

This chip allows for using 3.3V as the Vref or using an internal 2.4V source, though it seems that it's slightly better to use 3.3V:

I think their "peak-to-peak" value is equivalent to "noise-free bits". 

The smallest FSR is 3.3/64 = 0.05156V = 51.56mV.

Obtainable resolution

With these options known, I can calculate the smallest force I should expect to measure without any noise.

ADS131M08 24-Bit Simultaneous: 
    500SPS:
        500g * (9.375 / 3.3) / (2 ^(15-2.7))
        = 0.28168g
    1kSPS:
        500g * (9.375 / 3.3) / (2 ^(14.6-2.7))
        = 0.37168g
    2kSPS:
        500g * (9.375 / 3.3) / (2 ^(14.2-2.7)) 
        = 0.49044g

MCP3564R 24-Bit Fast Multiplex:
    1.6kSPS, 3.3V reference:
        500g * (51.56 / 3.3) / (2^13.8)
        = 0.54772g
    1.6kSPS, 2.4V internal:
        500g * (37.5 / 3.3) / (2^13.4)
        = 0.52564g

ADS1256 24-Bit Multiplex:
    800 or 400SPS (identical performance):
        500g * (78.125 / 3.3) / (2^14.4)
        = 0.54754g

ADS1262 32-Bit Multiplex:
    800SPS:
        500g * (78.125 / 3.3) / (2^12.5)
        = 2.04348g
    400SPS:
        500g * (78.125 / 3.3) / (2^16)
        = 0.18062g

 It seems that any option I pick would be able to get 0.5 grams of precision. Yes, this doesn't sound anywhere near as good as the 0.1g precision scales, but remember that I need this data correct in fractions of a second, not an entire second. 

Obviously, correct to the nearest 2 grams isn't going to work for a layout that is broken up into 8 gram layers, but if I can make do with 400SPS, I should expect to improve precision by over a magnitude. It is for this reason that -- PCB area permitting -- I'm probably going to go forward with the 32-Bit ADS1262 ADC.

Conclusion

This concept took 5 days to research and 6.75 hours to write out this log. The good news though is that Tetrinsic would now potentially be 

• more visually and dimensionally customizable (due to the 3D printed belt)
• coggless
• cheaper
• self-sterilizing 
• more tolerant on finger misalignment (since it's now a wide 10mm belt instead of the valley created between 2 ball-chains)
• more resilient to part sourcing issues
• better in terms of active-length to footprint-length ratio
• less software-complex since I don't need to communicate between 5 microcontrollers
• lower power since I don't have 5 microcontrollers that need power

whilst addressing the concerns Me In The Past had about this solution strategy, such as ease-of-manufacture and electronic restoring force.

[09:30, 28 Aug] 

It first helps to define what Tetrinsic needs to be at this point in time. It needs to:

1. Be no wider than 18mm.
2. Use a 45mm load cell or some other precision force sensing technology.
3. Precisely detect finger position.
4. Provide a grippable, low resistance, linear sliding medium for the fingers and thumb, ensuring fingernail compatibility.
5. Have a brushless motor that exerts at least 300gf at point of finger contact.
6. Allow for the below sliding medium configuration, unless another suitable ergonomic solution is found.

I've looked around AliExpress and there are a few motors that have a diameter of 16mm and lengths of 23mm and over.

Of which, the one with the best KV is this 1632 motor:

Specification:

Voltage range: DC 8V-18V.
Speed: 5600RPM-12000RPM.

Motor length: 32mm.
Motor diameter: 16mm.
Shaft diameter: 2mm.
Shaft length: 7mm.
Weight: 23g

Precision 16mm brushless motor with inner rotor, body length 32mm, the appearance of the bulk motor is somewhat scratched by transportation, the motor is relatively good quality, all aluminum alloy shell CNC processing, motor front and rear ball bearings, built-in Hall-type brushless motor.

The rotor is composed of 4 rare earth neodymium magnets. The magnetic force is very strong. It is obvious that the reluctance of the motor shaft is rotated manually. This motor does not have a drive, so it needs an external brushless drive to run.

The motor should be nominally used at 12V, and the measured voltage can also be used between DC 8V-18V.

Test data:
Voltage: 8V       No-load current: 0.08A       No-load speed: 5600RPM.
Voltage: 12V       No-load current: 0.13A       No-load speed: 8200RPM.
Voltage: 18V       No-load current: 0.16A       No-load speed:12000RPM.

The RPM at 12V is 17% higher than the 580KV motor. The KV range of this inrunner motor is 700 - 667KV. It also sounds like the cogging torque of this motor is notable.

I then watched the following:

The summary:

• Matching KV to the application is the most important
• Generally, inrunners have a higher KV than outrunners
• Inrunners can be more efficient as the windings can conduct heat through the case
• Cogging happens in a slotted but not slotless motor [relevant video]
• Cogging is stronger on a 6 pole motor.

Looking in the SmartKnob discord, it seems that for a device which is supposed to dynamically create it's own dedents, cogging is understandably not toleratable. This is what the creator means when he says that a motor has no cogging:

Even if that 1632 motor didn't sound like it had cogging so strong that the seller thought to mention it, I currently can't imagine a geometric configuration that I find acceptable.

Have you ever had that time when you're idly thinking and the thought "I should do something about that one thing before a problem arisises?". Well this morning I remember thinking 

You know, if those motors go OOS (out of stock) like what happened with SmartKnob, it's game over, right? Like... just saying... we should finalise this design fast or even just buy 20 motors now.

before replying to my other self

They've been on the market for like over a year at this point. I think we're fine.

Well, I was scrolling though this lovely looking website of the Ratchet H1, which is the first commercial haptic knob I've seen, and was thinking "Ok, maybe I was a bit too hard on myself complaining that the £25-or-so Tetrinsics were too expensive, considering this is almost 6X the price."

Then I remembered that I hadn't updated the BOM spreadsheet since Concept3.2 over a month ago.

First item: LCD screens.

Me: Oh, that's no longer in here. Delete that row.

Second item: 580KV Motor.

Me: Oh yeah I better go check to see that the price is still the same.

First link: £1.95

Me: !

Second link: £2.54

Me: ! !

The message over the basket: Sorry, this item is no longer available!

Me:

[  !  ]

AliTools search: Nothing remotely like the motors

The music I'm now playing in the background: Thunderbirds Are Go - Chaos Crew Theme (a fast paced, orchestra boss music track)

Me: Starts scrolling my wishlist where I certainly saved the motor

The background music when I manually get to the bottom of the page, seeing nothing: Quiet and ominous [00:47s]

Vivaldi: 0 search results for "BLDC" or "580KV"

Bing: Shows 4 results, all of them wrong

Google: Links me to the £2.54 listing 

Sigh. And just when I was feeling a bit excited with some "we are so back" energy. I was already kind of worried that these small motors were just old stock from something random. I've also read my fair share of products that have manufacturing or sourcing issues (like almost every hardware Kickstarter, Prusa, even Apple with the Vision Pro).

If the 580KV motors don't magically come back, or an alternative motor (I somehow missed every time I searched for it over the past 12 months) isn't discovered,

it is so...

over.

I first mentioned this in this Tetrescent log, but I've spent 3hrs modelling a new solution. The main reason being that most Aliexpress listings for stainless steel rods start at 2mm:

Secondly, even though one is probably never going to push down 500gf onto any Tetrinsic when in normal use (my thumb usually hits 420 - 450gf max and 300gf when not putting in the extra effort), the safety rating of 2.0 isn't ideal. 

It turns out that the solution to the question "Should I have the rods go over or under the collector?" is actually "Both.":

This is to allow the collector to obtain minimum wall distances of 1.2mm whilst also allowing the ball chain to slide past. I've also added this smooth curve to ease the transition from air to the side of the pressure collector:

The new assembly has a much nicer looking displacement profile:

The minimum safety rating is over 3, but it's unexpectedly in the area just under the load cell: 

Now that I've slept, this could be caused by the rod-ending offsets being too close together, meaning there isn't enough overlap where all 3 rods can bear the force. Increasing this offset to 27mm (from 20mm) reduces displacement to 0.300mm (from 0.324) and increases the safety factor to 3.78 (from 3.20). Reducing it to 10 increases displacement to 0.359mm and reduces the safety to 2.78:

I'll go with a 29mm offset to leave about 20mm from the last bend:

I'm now trying to see if I can actually obtain simulations for multiple force points by making the ones that have no force applied "frictionless":

No, that failed:

I'll just have the forces without the "frictionless" constraint. I'll assume that the extra bodies have a negligible weight.

It seems that Fusion does each solve sequentially. At least that means I could be inspecting one study while another one solves on the cloud. Anyway, here are the results:

I've set the displacement to "Actual" above, but to better see what's happening on a micron level, "Adjusted" is below:

It seems that the "minimum of minimum safety factors" is 2.94, when the force is applied near the ends of Tetrinsic. The outer force position has a max displacement is only 0.35mm, so it should feel rather solid throughout, even when pushing down at 500gf.

So I'm considering a 125mm solar cell for Tetent, and since this is much longer than my initial assumtion of a 75mm Tetrinsic, I thought I'd actually get a simulation.

I've modelled the tubes with a 60 degree inwards bend (instead of a 90 degree straight-down bend) so that the ball chain can take a shallower angle from the sprocket to the sides of the "pressure collector" as I call it. 

I've got a 500gf load exerted from the middle of the area, and it's already not looking so good, acheiving a minimum safety factor under 2. The good news is that, as expected, the load cell has almost no displacement, meaning that the force sensing area is symmetrical when under load.

The displacement is rather high, at 1.15mm. My goal is to cut that in half somehow.

This is where the outer tubes have their cutout not in the centre, which aims to have at least 2 tubes providing stiffness across the entire collector. I planned to do this on the 90mm Tetrinsic, but there wasn't enough length on the ends (only 5mm) before it needed to bend. Even still, this solution would be geometrically invalid since the tubes would intersect the countersunk screws.

Unfortunately, the displacement only improves by 0.1mm and the safety factor is unchanged.

Returning to the aligned-end tubes and having them made from solid steel (so they're more like 1.5mm thick wires), I get a displacement of 0.86mm and a safety factor of 2.

Interestingly, the displacement is still 0.86mm when the force is applied directly under another Tetrinsic, but the right side is greener than the left side. Green means over 0.4mm of displacement.

The displacement is more balanced with the uncentered cutouts. I probably should go with the other solution, whereby the collector is on the "outside" of the loop.

So this is it. It certainly looks more complex. The main benefit of this strategy is that the finger force pushes the tubes further into the collector, instead of pushing the tubes out of it. 

Another benefit is that, similar to a boat through water, I can print a seperator that splits the dual chain.

With this, I get a max displacement of 0.77mm and the base has a displacement of 0.34mm.

Considering print orientation, it's probably best if this face was horizontal instead of perpendicular to the tubes. 

it seems that a height offset of 5.2 - 5.5mm would be required:

As will M3x10 countersunk screws:

So these bolts were actually causing the simulation to fail until I did a combine-cut to eliminate the geometric intersection between them and the load cell.

Now, for some reason, trying to simulate forces from distances over the 9.3mm set in the above simulation caused the simulation to fail. I tried a few different things and red the error logs, and I've been able to get a 25mm force sim to succeed by reducing the mesh setting to around 7%:

I simulated 50mm, and for some reason, only the "Displacement" view is glitched:

45mm worked though:

It does seem that I should also have a 0.75mm spacing below Tetrinsic to allow for this part to flex uninterrupted.

It's not much, but it should give you a slightly better idea on how everything fits together. It seems that I've sketched the mirrored version of what I intended.

Also note that the bottom of the steel tubes are slightly closer together than the top, This is to give the ball-chains more space. The angle is only 2 degrees, which isn't significant enough to have 2 slightly different bending jigs. It's possible that I increase the angle should I need more space, but right now I've kept a 1.2mm gap for the walls of the printed part that holds these in place. The tubes also act as reinforcement for this printed holder.

Since this new Tetrinsic now has 3D geometric considerations to uphold, I was having trouble thinking about the solution. Thus, I decided to create some rough sketches of the problem in Paint3D. Being able to digitally sketch to help solve problems is the main reason behind the Sketchθ mode in Tetent.

Initial Drafting

Basically, it was a bunch of drawings to figure out what could and couldn't go where, as well as making it easier to see how design decisions would propogate.

Once it seemed that I had an idea that could work, I continued the sketch in Fusion360 where I could get accurate scaling. It seems that 4.8mm is a good starting point as a Tetrinsic offset. I'm also planning for space for a 4.1 inch HMI, since it's almost the same dimensions as the panel by itself with the main difference being thickness.

As explained in this log, I believe the HMI is the best solution but it's also the one that uses the most space, thus if I can accomodate this display, I could later decide on going with any of the other solutions discussed instead.

In the free space available, I've found this 356575, 2.5Ah battery:

To allow for enough space, I've gone with 17 teeth sprockets. The total height is thus 27.5mm. That's the same height as the TestTetrinsics I printed.

Starting yesterday, I've been conceptualising a way to get a fifth Tetrinsic to fit. 

One of the ergonomic issues with Thumb1 is that it and Finger5 are both what I'd call "aligning fingers". As one would expect, the solution ergonomically fails if only one of them is in alignment. That's part of the reason why the main Tetent concepts have only used 4 fingers:

Unfortunately, any ergonomic solution failed geometrically due to the size of Tetrisic. #Tetent TestCut [gd0139] is no longer geometrically viable with how large Tetrinsic has grown now.

However, with this new idea that I talked about in the previous log, I feel that there's a possibility of obtaining one of the best ergonomic solutions possible whilst still being ergonomically viable, though it could be harder to manufacture since a closed loop would form around the FingerN Tetrinsics. The good news is that I can make it ambidexterous without needing an additional Tetrinsic, now that it's so long.

This strategy actually goes all the way back to this log here where there's even a handy diagram. Right now, I'm targetting a height offset of 6mm. Conveniently, the 1.5mm stainless steel tubes arrived. 1 tube alone already feels unmoving, so the 3 tubes that make up the sliding surface of Tetrinsic should be stiff enough for the task.

I'm also removing the LCD/Solar stuff from Tetrinsic. I'll still be keeping compatibility by exposing pins on Tetrinsic PCB, but it now seems that the aesthetic should be determined by the device that Tetrinsic is implemented in. For example, why have a small artisan keycap when the entire free area could be one, such as a topology map:

At the moment, I'm expecting the visible area to be about 75mm x 75mm for Tetent. This approach to lighting has come full circle, as the Tetwin Switches were designed to be transparent so that an external light source could shine though.

So, unfortunately, with the expected size of Tetrinsic, it's looking like a possible Tetent solution would be very similiar in dimensions to an AirBerry (my Let's Split keyboard), just rotated 90 degrees such that it's 6 rows, 4 columns instead of the other way around. I was imagining something like this:

It's... fine... simple yet modern... but it's not exactly the kind of device I hoped to get out of 19 months of R+D work.

However, just after writing about the linear motor research in the previous log, I thought of what a new Tetent could look like with one and thought of sometihng similar to this:

But then I remembered that a) the ball-chain of the current Tetrinsic runs on a stainless stainless steel tube, b) that tube is probably stiff and c) that a good part of the proposed Tetrinsic is empty space on the bottom half. Thus, why not turn the concept up-side down and bring back some level of future-modern, Apple-esque aesthetic that I'd expect to see on Yanko Design?

The most notable benefit of this is that a larger, single display could be used. However, as a larger screen means more pixels to drive, I'll most likely be sticking with the 2x 1.14" LCDs. Additionally, the only OLED I could find, the 3.8" one I was planning to use for #Tetent TimerSpy [gd0136], actually has poor black levels in ambient light. I can clearly see the difference between "black" on the OLED and the black bezels from this video

Another notable mention is that this extends to any material, such as a photovoltaic cell or a porcelain tile. Lastly, a benefit is that the pressure plates won't be right next to each other, meaning that I could actually have the Tetrinsics closer without worrying about manufacturing tolerances.

The next Tetent concept is currently a work in progress, but this sketch (which is one half) should give an idea as to the strategy I want to try. For starters, I'm thinking that the squiggly bit is where I could have a fancy design on the outside. Behind that, on the inside, it's where the load cell, battery and top screen would be. The square (highlighted below) is where the memory LCD would go, and the motors behind it. 

Hopefully, this strategy makes the resulting Tetent designs look more futuristic, even if I might not be able to make Tetrinsic as thin as I'd like.

The 3D of the above looks like this, though I'll post updates on the overall design in the Tetent project:

At this time of writing, I'm still going to go thorugh with the soon-to-be-designed dual motor Tetrinsic, but I'm still not happy about the length and height of the current solution. While the active length of 63mm makes sense as all 4 fingers can ergonomically exist in the resulting quad-Tetrinsic area, it only accounts for 54% of the total length. The thickness of 28-29mm is also limiting, and ideally Tetrinsic would be half this.

Other than width, the main ergonomic issue I need to engineer for is to obtain reliable fingertip grip whilst having a very low friction sliding surface, even when pressure in excess of 100gf per finger is applied.

To have a chance of meeting an under 15mm thickness target, a potential solition is to give up infinite-scroll and off-the-shelf hardware for a custom linear motor:

Unfortunately, I think the bigger issue is developing the software for Tetrinsic and mounting strategies, so right now I need a development solution and not an ideal solution.