TM1814 vs TM1812 | RGBW vs RGB LED Driver IC Comparison

TM1814 and TM1812 share a family name and a single-wire interface, and almost nothing else. TM1814 is a four-channel adjustable-current driver for one WRGB pixel in a SOP8. TM1812 is a twelve-channel fixed-current driver for four RGB pixels in a SOP16. Buyers arrive from two directions: a decorative fixture that needs a real white channel, or a four-pixel module where the board is already tight and the firmware already exists.

Frame length and current architecture differ, so this is not a matter of preference. Choose wrong and the strip either rejects your controller’s data or runs hot enough to lose white-point consistency across a facade. Below are the five points that decide the order — frame format, current setting, package, cascade depth, thermal headroom — followed by how to verify a lot before committing to a reel.

Two parts, two jobs

TM1814 has four constant-current sinks, one per colour plus white, so one device drives one pixel. What matters is the current stage: each output is trimmed in 64 steps from 6.5 mA to 38 mA by a command word sent ahead of the pixel data, so white can be trimmed differently from red, green and blue without a BOM change. Output withstand is 32 V, VDD carries an internal 5 V regulator so a 6–24 V rail needs only a series resistor, and current error is ±3 % channel-to-channel and ±5 % chip-to-chip.

It also has a second personality. If DIN stops for roughly 500 ms, TM1814 enters internal-control mode and runs a fixed ten-step white/red/green/blue sequence while still forwarding data downstream. For a rental screen that must keep doing something after the controller dies, that is a feature. For an installation that should go dark on fault, it is behaviour you must design around, because there is no way to switch it off.

TM1812 is the density part: twelve outputs arranged as four RGB groups, 24 bits per pixel, 45 mA typical per channel, no trim command. It is specified at 800 kbps with no speed-select pin, holds at least 1,080 pixels at 30 fps, and comes in SOP16 and DIP16. There is no internal pattern mode, so a failed controller leaves the string holding its last state or blank rather than cycling colours.

TM1814 vs TM1812: side-by-side specification

Parameter	TM1814	TM1812	What it means for the design
Channels	4 (W, R, G, B)	12 (four RGB groups)	One pixel per TM1814, four per TM1812
Bits per pixel	32 (8 per channel)	24 (8 per channel)	Different frame length — firmware is not portable
Command overhead	C1 + C2, 64 bits per frame	None	TM1814 needs a current word and its inverse every frame
Channel current	6.5–38 mA, 64 steps	45 mA typical, fixed	TM1814 trades brightness against heat; TM1812 does not
Current accuracy	±3 % channel, ±5 % chip	Per-channel figure from the datasheet	White-point consistency across a long run
Grey scale	256 levels per channel	256 levels, finer at low brightness	Both adequate for video-rate dimming
Output withstand	32 V absolute, 24 V recommended	30 V	Both suit 12 V and 24 V rails
Logic supply	4.5–6.5 V, 5 V typical	5 V class, internal regulator	Neither is a 3.3 V part
Data rate	800 kHz external, 500 kHz internal mode	800 kbps, fixed	Same nominal clock, different frame
Internal pattern mode	Yes, after ~500 ms without DIN	No	Fail-visible or fail-silent, by part choice
Cascade depth	1,000 pixels at 25 Hz, 800 at 32 Hz	≥1,080 pixels at 30 fps	TM1812 goes deeper on a single run
Package	SOP8, 1.27 mm pitch	SOP16, DIP16	DIP16 still matters for repair stock
Dissipation	250 mW at 25 °C	Package-dependent	Four channels in a small body is a thermal question
Operating temperature	−40 to +85 °C	−40 to +85 °C	Neither is automotive-qualified

The data frame decides compatibility, not the channel count

A TM1812 frame is a flat sequence of 24-bit words. The first device keeps 24 bits and forwards the rest; the next keeps the next 24. Any controller with a 24-bit-per-pixel loop drives it, which is why the part is everywhere in low-cost RGB modules.

TM1814 is not that. A frame is C1, C2, one 32-bit PWM word per device, then a reset pulse. C1 holds four 6-bit current values — white, red, green, blue — with the top two bits of each byte fixed at zero. C2 is C1 with every bit inverted, and the device validates the pair at reset: if C2 is not the exact inverse, the new current is not applied. Every device forwards C1 and C2; each keeps the first 32-bit word that follows them.

Three consequences cost money if you meet them after tooling:

Firmware is not portable. A 24-bit library cannot drive TM1814 at the same 800 kHz. You need a 32-bit word with white first, plus a current command and its inverse at the head of each frame.
One string, one protocol. Mixed devices shift every downstream pixel by eight bits and produce scramble, not a colour error software can correct.
There is a fixed 32-bit offset. What a device forwards always trails what the controller sent by 32 bits, and the command pair adds 64 bits per frame — about 80 µs at 800 kHz. Account for it in diagnostics or you will chase a fault that does not exist.

Current setting: 64-step trim against a fixed 45 mA

A fixed 45 mA channel forces the white LED to be matched by other means: a series resistor on W, a different W bin, or a white point that drifts as the fixture warms. On a 100-pixel string that is four hundred resistors and a tuning step in production.

TM1814 puts the trim in firmware. White at 12 mA with red, green and blue at 30 mA, and the same reel re-tuned for a new bin without a layout change. For a buyer running one board across several W bins, that is the difference between one approved BOM and one per bin.

The accuracy figures do the rest. ±3 % channel-to-channel and ±5 % chip-to-chip is what stops a long facade showing steps between nodes at low grey levels. Ask for that figure in writing on any quote: it is what separates a genuine Titan Micro reel from a substitute that measures well on a bench and fails on a wall.

Package, pin count and what it costs on the board

Four RGBW pixels need four TM1814 or one TM1812: four SOP8 bodies against one SOP16. The trade is placement cost and board area against device count and reel handling. Both use a 1.27 mm pitch, so stencil and reflow profile are unchanged. Two details are easy to miss. TM1814 has no DIP option, which matters to a repair bench that has always stocked DIP16 TM1812. And four output transistors plus a regulator on a small pad area concentrate heat where a SOP16 spreads it, so on a dense strip without copper pour the effect shows as a gradient along the run rather than one hot spot.

Cascade depth and refresh budget

At 800 kHz a bit takes 1.25 µs, so a 32-bit word takes 40 µs. Titan Micro publishes the TM1814 limits: 400 pixels refresh in 16 ms (62.5 Hz), 800 in 32 ms (32.2 Hz), 1,000 in 40 ms (25 Hz). TM1812 is specified for at least 1,080 pixels at 30 fps, roughly a third deeper on one run.

A 100-node RGBW installation is comfortable: 4 ms of payload plus 80 µs of overhead, well inside 100 Hz. The failure at this depth is rarely bandwidth but the inter-byte gap — idle more than 50 µs between words and the device may read a reset, drop the frame and restart. Software-bitbanged drivers on a loaded MCU do this constantly, and the symptom, the first pixels updating while the rest flicker, usually gets blamed on the reel.

Power, series resistors and the 1.6 V dissipation ceiling

TM1814 runs from 5 V directly, or from 6–24 V through a VDD resistor sized at (supply − 5.5 V) ÷ 10 mA: 50 Ω at 6 V, 350 Ω at 9 V, 650 Ω at 12 V, 1.8 kΩ at 24 V. Fit 100 Ω in series with DIN and DO for hot-plug transients, and 100 nF at every device with the shortest possible trace. Skipping that capacitor is the usual reason a string works on the bench and fails in a metal housing.

The output stage is where designs go wrong. Constant current starts at about 0.8 V across the output, and Titan Micro recommends 1.2–3 V. Higher is not better: everything above the LED forward voltage is dissipated in the package. Four channels at 38 mA sink 152 mA, and against 250 mW at 25 °C that budget is spent at roughly 1.6 V average.

A real case: a 12 V rail with three LEDs in series at about 9.6 V forward leaves a 2.4 V drop — 91 mW per channel, 365 mW per device, over budget. At 20 mA it is 192 mW and holds. A fixed-45 mA part cannot make that trade: twelve channels at 45 mA into the same drop would carry well over a watt. That is why 12 V TM1812 designs keep the LED string close to the rail, and why a 24 V TM1814 design should not assume the maximum setting is available.

Which part fits which product

The part travels with the fixture rather than the market, but enquiries differ by region. Indian decorative and street-furniture builders take TM1814 for RGBW pole and facade nodes, TM1812 for four-pixel modules. Vietnamese and Thai strip assemblers buy both in reel quantities for export programmes; Malaysian signage shops lean on TM1812 because their controllers are already 24-bit. Middle East facade contractors specify 24 V and ask about ambient derating first, while European and US stage buyers want the datasheet revision and RoHS paperwork before price.

12 V RGBW strip, white channel required — TM1814; the per-channel trim is the reason to choose it.
24 V architectural node, one to three LEDs per channel — TM1814 with the 1.8 kΩ resistor and a current setting chosen against the drop, not the maximum.
Four-pixel RGB module on a proven 24-bit controller — TM1812. Rewriting working firmware to gain a channel you do not need is the wrong trade.
Signage that must keep showing something after a controller failure — TM1814, if the client accepts a fixed colour cycle instead of their content.
Installation that must go dark on fault — not TM1814. Use TM1812 with a hold-or-blank policy in the controller.
One high-power LED per node, 100–150 mA — TM1814 with all four outputs paralleled. Each adds 38 mA, so four give 152 mA, and the firmware must write all four registers for the current to split evenly.
More than 1,000 pixels at 30 fps — TM1812, or split TM1814 across two data outputs.
Segment display, appliance panel or LCD glass — neither. That is TM1637, TM1650, TM1628 or TM1621B territory.

Price and availability: what actually moves the quote

Both are commodity devices with thin margins, so a wide spread between quotes is a signal, not a bargain. What moves the number: die and test yield in the current window, reel quantity, date-code age, and whether the quote covers original Titan Micro product or something the seller calls equivalent. On small orders, freight and incoterm matter more than buyers expect.

The structural problem is that Titan Micro sells domestically and has no English-language export desk. An overseas search for TM1814 stock usually ends with a catalogue reseller applying a retail multiplier, or a broker offering a partial reel with mixed date codes. Neither is a supply chain.

Working with Newfly: the numbers

Buyers evaluating a supplier want the commercial parameters stated, not implied. Here is ours. Where a figure depends on your quantity, package or destination, we say so rather than publishing a number that will not hold.

Item	What we commit to
Sample order	Sample and small-batch quantities are supported. Sample fee, freight terms and dispatch time are confirmed in writing with each quotation.
Minimum order	Flexible. We support partial-reel and small-batch quantities for first orders and scale with you.
Lead time	Confirmed in writing per order, against the quantity and destination you specify.
Inspection	Marking, packaging and date-code verification before dispatch. Tell us your acceptance criteria and we will align to them.
Defect claims	The defect resolution process is agreed in writing before the order, not after a problem appears.

TM1814: where it does not belong

Specification tables tell you what a part can do. Design failures come from what it cannot do. Before you commit a TM1814 to a production board, check these cases — if any of them applies to your design, this is the wrong part and we will say so.

your panel is a segment display or an appliance control board — TM1814 is a pixel driver with no digit decode, no display RAM and no key scan. Use TM1637 or TM1650 for a 16-pin segment driver, TM1628 or TM1668 where you need more digits and keys, and TM1621B, TM1621C or TM1621D for LCD glass.
your MCU rail is 3.3 V and you intended to drive DIN from it directly — the minimum input high level is 3.5 V at a 5 V supply, so a 3.3 V rail sits below threshold. Add a level shifter or run the logic at 5 V. This is the most common mis-order on both parts.
your controller is a fixed 24-bit loop written for TM1812 — TM1814 needs a 32-bit word per device plus the C1/C2 current command and its inverse. A 24-bit library will produce scramble, not a colour error you can trim out in software.
the installation must go dark when the controller fails — TM1814 enters internal-control mode after roughly 500 ms without DIN and will cycle a fixed white/red/green/blue pattern for as long as it is powered. There is no disable. Use TM1812 and a hold-or-blank policy in the controller if silence is a requirement.
you need more than 38 mA on four independent channels at once — the four outputs can be paralleled to reach 152 mA, but then they are one channel, not four. If the design needs four independent channels above 38 mA, this part cannot do it.
your LED string forward voltage sits far below the rail — the difference is dissipated in the package. Four channels at 38 mA into a 2.4 V drop is 365 mW against a 250 mW rating at 25 °C. Reduce the current setting, shorten the string, or move to a part with an external switch.
your programme requires AEC-Q100 qualification or a formal PPAP package — these are commercial-grade parts specified over -40 to +85 °C with no automotive qualification attached. We will decline the quotation rather than imply a grade the part does not carry.

Buying guide: verify the lot, then the supplier

TM1814 is widely copied, and the copies pass a bench test because a constant-current sink is easy to imitate. What they do not imitate is the command layer. Four tests, all runnable on a ten-device sample:

C1/C2 validation. Send a valid current command with its inverted pair and confirm the measured current changes. Then send C1 with C2 not inverted: a genuine device leaves the previous setting in place, a copy that skips the check applies the new value.
64-step sweep. Set all-zero (6.5 mA) and all-one (38 mA), measure across a shunt. Identical readings mean the trim stage is missing, whatever the marking says.
Internal mode. Remove DIN for more than 500 ms. The device should start the ten-state white/red/green/blue sequence and keep forwarding to DO. A copy with no MCU block goes dark, which is visible in seconds.
Timing. Drive a 2.5 µs period with the 0 and 1 low times inside the published windows. TM1814 tolerates it; marginal silicon often does not. A failure is a reject, a pass is only a screen.

Then the supplier. Ask for the reel label and a photograph of the marking and date code, not a stock image of a generic SOP8, and confirm the lot is a single date code. Ask what is inspected before dispatch — marking, packaging, date-code verification — and whether third-party inspection is available. Ask for the documents your customs regime requires. And ask what the minimum order is, because a supplier who will not cut a partial reel for validation is telling you how they will handle a claim.

When we tell you not to buy

A supplier who accepts every enquiry without question is not screening anything. These are the situations where we will recommend against the purchase:

You are re-laying a board to TM1814 only to gain a white channel, and the existing TM1812 design already meets the specification with a resistor on W. We will show you the arithmetic before you spend on tooling.
The lot offered to you has mixed date codes on a single reel and no traceable origin. We will describe the risk rather than quote it as normal stock.
You want a full carton before the part is validated on your own board. We will say so before you commit working capital to an unproven design.
You are specifying the 38 mA maximum on a 24 V rail without checking the output drop. We will run the dissipation figure with you rather than ship a reel that runs at the edge of its package rating.
Your procurement rules require a franchised-channel paper trail that only the manufacturer’s authorised channel can issue. We are an export distributor, not the franchised channel. If that paper trail is mandatory for the line, we are not your supplier for it.
Your target price can only be met with a lot we cannot verify as original Titan Micro product. We will decline the order rather than ship an unverifiable lot.
You are asking us to quote WS2814, SK6812 or another vendor’s RGBW driver alongside TM1814. We supply the Titan Micro side only and will say so rather than imply stock we do not hold.

Why buyers source these parts through Newfly Electronics

Newfly Electronics supplies Titan Micro (天微电子) display and LED driver ICs to manufacturers, design houses and repair operations outside China, with TM1814 and TM1812 alongside the segment drivers. The advantage is structural: we buy in the same domestic market Titan Micro sells into, so the quote is a China supply-chain price rather than an export-desk price with a distributor layer on top.

The support is specific to these two parts: we will tell you which part your frame format can accept before you order either, send the current datasheet revision with the quotation, and answer the pinout and current-setting questions that arise in first-article builds. If you need a 3.3 V interface, a deeper cascade than one run allows, or an automotive qualification we cannot provide, we will say so rather than quote around it. Samples and partial reels sit alongside full-reel and scheduled production orders, with export packing and documentation handled from Shenzhen.

FAQ: TM1814 and TM1812 buying questions

Can TM1814 replace TM1812 on an existing board?

No. The frame length differs — 32 bits per pixel plus a current command against a flat 24-bit stream — and the package changes from SOP16 to SOP8. Firmware and layout both have to change, and a mixed string will scramble every downstream pixel.

How many TM1814 devices does a 100-pixel RGBW string need?

One hundred, one per pixel. The payload is 3,200 bits, or 4 ms at 800 kHz, plus 80 µs of command overhead — well inside a 30 fps budget. The practical limit on one run is around 800 pixels at 30 fps.

Is TM1814 still in production?

Yes, it is an active Titan Micro part and we ship current date codes. If a lot we hold is older than the current production window, we tell you the date code before you order rather than after.

Can I drive TM1814 from a 3.3 V microcontroller?

Not reliably. At a 5 V supply the minimum input high level is 3.5 V, so a 3.3 V rail sits below the threshold. Use a level shifter or run the logic at 5 V. The same applies to TM1812.

Why does my string start flashing colours on its own?

DIN has stopped for about 500 ms and the device has entered internal-control mode. Check the inter-byte gap first: an idle over 50 µs mid-frame can be read as a reset, so a controller that stalls trips the same timer.

Does TM1814 need a series resistor on VDD, and what package is it?

SOP8 only, on a 1.27 mm pitch, with no DIP option. A VDD resistor is needed only above 5 V: 50 Ω at 6 V through to 1.8 kΩ at 24 V, per the published table.

Does Newfly Electronics provide the datasheet and a sample?

Yes to both. We send the current Titan Micro revision with the quotation, and sample or partial-reel quantities can be dispatched for validation before a production order.

Request a Quotation

Send us the part number, the package and the quantity you need. We will reply with price, confirmed stock and lead time for the a TM1814 or TM1812 requirement.

Contact Newfly Electronics
Email: sales@newflyelectronics.com
WhatsApp: +86 15112635657