Recent homebrew has adapted the SNES mouse to be used with the NES. Hori released a trackball for Famicom called the Hori Track. A famiclone made by the Subor company also had a mouse. All three use different protocols.
The Super NES Mouse (SNS-016) is a peripheral for the Super NES that was originally bundled with Mario Paint. It can be used with an NES through an adapter, made from an NES controller extension cord and a Super NES controller extension cord, that connects the respective power, ground, clock, latch, and data pins.
As with the standard controller, the mouse is read by turning the latch ($4016.d0) on and off, and then reading bit 0 or bit 1 of $4016 or $4017 several times, but its report is 32 bits long as opposed to 8 bits.
On an NES or AV Famicom, the mouse may be connected to bit 0 through the front controller ports. On the original Famicom, it would normally have to be connected to bit 1 instead through the expansion port.
The report is divided functionally into four bytes. The most significant bit is delivered first:
76543210 First byte ++++++++- Always zero: 00000000 76543210 Second byte ||||++++- Signature: 0001 ||++----- Current sensitivity (0: low; 1: medium; 2: high) |+------- Left button (1: pressed) +-------- Right button (1: pressed) 76543210 Third byte |+++++++- Vertical displacement since last read +-------- Direction (1: up; 0: down) 76543210 Fourth byte |+++++++- Horizontal displacement since last read +-------- Direction (1: left; 0: right)
After the fourth byte, subsequent bits will read as all 1, though the Hyperkin clone mouse instead reads a single 1 then all 0s. 
The Hyperkin mouse will not give a stable report if it is read too fast. Between each bit read, there should be at least 14 CPU cycles. Between the 2nd and 3rd byte (16th and 17th bit) of the report should be at least 28 CPU cycles. Reading faster than this will result in corrupted values..
Motion of the mouse is given as a displacement since the last mouse read, delivered in the third and fourth bytes of the report.
The displacements are in sign-and-magnitude, not two's complement. For example, $05 represents five mickeys (movement units) in one direction and $85 represents five mickeys in the other. To convert these to two's complement, use negation:
; Convert to two's complement lda third_byte bpl :+ eor #$7F sec adc #$00 : sta y_velocity lda fourth_byte bpl :+ eor #$7F sec adc #$00 : sta x_velocity
When the magnitude of motion is 0, the reported sign will repeat the last used sign value for that coordinate.
The mouse can be set to low, medium, or high sensitivity.
On the original SNES mouse this can be changed by sending a clock while the latch ($4016.d0) is turned on:
ldy #1 sty $4016 lda $4016 dey sty $4016
Some revisions of the mouse's microcontroller reportedly power up in an unknown state and may return useless values before the sensitivity is changed for the first time.
The clone Hyperkin "Hyper Click Retro Style" mouse will not cycle its sensitivity this way. Instead it has a manual button on the underside that must be pressed by the user to cycle sensitivity. For this reason, it is not advised to use the software sensitivity cycling to automatically detect the presence of a mouse.
On the original SNES mouse, sensitivity setting 0 responds linearly to motion, at a rate of 50 counts per inch. Values range from 0 to 63, but values higher than 25 are increasingly difficult to produce. 
Sensitivity settings 1 and 2 appear to remap the equivalent setting 0 values 0-7 to a table, and clamping at the highest value. (Rarely, however, other values may be seen in settings 1 and 2.)
The Hyperkin's two manually selected sensitivities both scale linearly with motion speed. Low sensitivity produces 0-31, and high sensitivity produces 0-63. The magnitude of the result is not dependent on the rate of polling, so it appears to report the current speed rather than the distance travelled since the last poll. The maximum value (31/63) at either sensitivity appears to correspond roughly to a speed of 8 inches per second. (This mouse should be used on a surface with a visible texture.)
A program that reads the mouse while playing DPCM samples must proceed very carefully. Sample playback causes occasional double reads on $4016 and $4017, deleting a bit from the stream read back. The re-reading solution that can be used for the standard controller fails here because each latch pulse sent to a mouse will clear its count of accumulated movement, and there is no known way to reliably detect corrupted data. Only the OAM DMA sync trick reliably avoids these glitches.
Using more than two mice on an AV Famicom is not recommended for two reasons:
- A mouse draws 50 mA, which is much more current than the standard controller draws. Drawing too much current is likely to cause the voltage regulator to overheat.
- Changing player 1's sensitivity also affects player 3's, and changing player 2's sensitivity also affects player 4's.
Some documents about interfacing with the mouse recommend reading the first 16 bits at one speed, delaying a while, and reading the other 16 bits at another speed, following logic analyzer traces from a Super NES console. However, these different speeds are merely an artifact of the main loop of Mario Paint, and the mouse will give a correct report when read at any reasonable speed. For example, a program could read 8 bits, wait a couple thousand cycles, and then read the other 24.
Hori produced a trackball compatible with Moero Pro Soccer, Moero Pro Soccer, Putt Putt Golf, and US Championship V'Ball. It was released in Japan, and what appears to be a prototype U.S. version was exhibited behind glass in Nintendo World, but the U.S. version never reached stores.
Report byte 1 is the embedded standard controller.
Byte 2, MSB first:
7654 3210 |||| ++++- Axis 2, signed 4 bit, XOR with $F ++++------ Axis 1, signed 4 bit, XOR with $F
Byte 3, MSB first:
7654 3210 |||| ++++- Unknown (read and unused by games) ||++------ ID byte (1 or 2 depending on version) |+-------- Unknown (speculated by nocash to be the speed switch) +--------- Rotation mode switch (0: R, 1: L)
In rotation mode L, Up on the Control Pad points up, axis 1 points down, and axis 2 points right. In rotation mode R, Up on the Control Pad points right, axis 1 points left, and axis 2 points down.
This came with Subor's SB2000 famiclone system.
- Educational Computer 2000
Input ($4016 write)
7 bit 0 ---- ---- xxxx xxxS | +- Strobe
Output ($4017 read)
7 bit 0 ---- ---- xxxx xxxD | +- Serial data
The mouse returns either 1-byte or 3-byte responses based on the size of the x/y axis movement since the last read. If the movement value for both the X and Y axis is between -1 and 1 (inclusively), the mouse returns a 1-byte response. Otherwise, a 3-byte response is sent. In this case, the strobe bit must be toggled 3 times in a row to read the entire response (e.g read $4017 8 times, toggle strobe bit, read 8 more times, etc.)
1-byte response format
LRXXYY00 |||||||| ||||||++- Always 0 ||||++--- Y movement (0: no movement, 1 or 2 = went down 1 unit, 3 = went up 1 unit) ||++----- X movement (0: no movement, 1 or 2 = went right 1 unit, 3 = went left 1 unit) |+------- Right mouse button (1: pressed) +-------- Left mouse button (1: pressed)
3-byte response format
LRSXTY01 |||||||| ||||||++- Always 01 |||||+--- Y movement (Bit 4) ||||+---- Y movement direction (1: up, 0: down) |||+----- X movement (Bit 4) ||+------ X movement direction (1: left, 0: right) |+------- Right mouse button (1: pressed) +-------- Left mouse button (1: pressed)
--XXXX10 |||||| ||||++- Always 10 ++++--- X movement (Bits 0-3)
--YYYY11 |||||| ||||++- Always 11 ++++--- Y movement (Bits 0-3)
- forum post: Hyperkin SNES mouse investigation
- forum post: Hyperkin mouse reads have a speed limit
- forum post: Hyperkin SNES Mouse cannot software-cycle sensitivity
- FullSNES - Nocash SNES Mouse documentation
- forum post: SNES Mouse sensitivity measurements
- forum post: Hyperkin Mouse sensitivity measurements