45 MicroTapeout (of sky130 cells)

45 : MicroTapeout (of sky130 cells)

Design renderSelect Project123456789ONSW2
  • Author: htfab
  • Description: 395 standard cells with a mux to select between them
  • GitHub repository
  • Clock: 10000 Hz

How it works

Digital chip designs are usually written in a hardware description language like RTL Verilog and then synthesized into a set of mask layers suitable for fabrication. In order to make both synthesis and verification robust for huge designs, a modular approach is used where the functionality of the circuit is decomposed into pre-built blocks called standard cells with well-known and thoroughly tested behaviour and layout.

This design contains a copy of most standard cells in the sky130_fd_sc_hd library along with a multiplexing mechanism that allows exposing any of them to the input/output pins.

An MPW shuttle fabricates multiple designs on the same wafer. TinyTapeout merges several projects in a single shuttle submission. MicroTapeout pushes the limit with each block containing just a single cell. Apart from the geek factor the fabricated chip can be used by low-level digital design engineers to better understand the behaviour of the individual standard cells and might even provide some timing insights.

There are 437 standard cells in our library, of which 42 don't produce output or require special power handling. This leaves us with 395 cells. Each cell has up to 6 inputs and up to 2 outputs for a total of 427 outputs. The same 6 inputs are fed into each cell in parallel while the 427 outputs are divided into 54 pages of 8 outputs each with a multiplexer deciding which page is mapped to the output pins.

In order to drive the 6 cell inputs and the 6 bits of input to the mux from a total of 8 input pins we use some registered logic. Input pin 0 is a clock signal while input pin 1 selects page mode. On each rising clock edge we save input pins 2 to 7 into a page register if page mode is on and into an input register if page mode is off. Cell inputs are then supplied from the input register and the mux operates on the page register.

Mapping of outputs to pages:

page pin pin 0/4 pin 1/5 pin 2/6 pin 3/7
000000 0-3 conb_1.h conb_1.l buf_1 buf_2
4-7 buf_4 buf_6 buf_8 buf_12
000001 0-3 buf_16 bufbuf_8 bufbuf_16 inv_1
4-7 inv_2 inv_4 inv_6 inv_8
000010 0-3 inv_12 inv_16 bufinv_8 bufinv_16
4-7 and2_0 and2_1 and2_2 and2_4
000011 0-3 and2b_1 and2b_2 and2b_4 and3_1
4-7 and3_2 and3_4 and3b_1 and3b_2
000100 0-3 and3b_4 and4_1 and4_2 and4_4
4-7 and4b_1 and4b_2 and4b_4 and4bb_1
000101 0-3 and4bb_2 and4bb_4 nand2_1 nand2_2
4-7 nand2_4 nand2_8 nand2b_1 nand2b_2
000110 0-3 nand2b_4 nand3_1 nand3_2 nand3_4
4-7 nand3b_1 nand3b_2 nand3b_4 nand4_1
000111 0-3 nand4_2 nand4_4 nand4b_1 nand4b_2
4-7 nand4b_4 nand4bb_1 nand4bb_2 nand4bb_4
001000 0-3 or2_0 or2_1 or2_2 or2_4
4-7 or2b_1 or2b_2 or2b_4 or3_1
001001 0-3 or3_2 or3_4 or3b_1 or3b_2
4-7 or3b_4 or4_1 or4_2 or4_4
001010 0-3 or4b_1 or4b_2 or4b_4 or4bb_1
4-7 or4bb_2 or4bb_4 nor2_1 nor2_2
001011 0-3 nor2_4 nor2_8 nor2b_1 nor2b_2
4-7 nor2b_4 nor3_1 nor3_2 nor3_4
001100 0-3 nor3b_1 nor3b_2 nor3b_4 nor4_1
4-7 nor4_2 nor4_4 nor4b_1 nor4b_2
001101 0-3 nor4b_4 nor4bb_1 nor4bb_2 nor4bb_4
4-7 xor2_1 xor2_2 xor2_4 xor3_1
001110 0-3 xor3_2 xor3_4 xnor2_1 xnor2_2
4-7 xnor2_4 xnor3_1 xnor3_2 xnor3_4
001111 0-3 a2111o_1 a2111o_2 a2111o_4 a2111oi_0
4-7 a2111oi_1 a2111oi_2 a2111oi_4 a211o_1
010000 0-3 a211o_2 a211o_4 a211oi_1 a211oi_2
4-7 a211oi_4 a21bo_1 a21bo_2 a21bo_4
010001 0-3 a21boi_0 a21boi_1 a21boi_2 a21boi_4
4-7 a21o_1 a21o_2 a21o_4 a21oi_1
010010 0-3 a21oi_2 a21oi_4 a221o_1 a221o_2
4-7 a221o_4 a221oi_1 a221oi_2 a221oi_4
010011 0-3 a222oi_1 a22o_1 a22o_2 a22o_4
4-7 a22oi_1 a22oi_2 a22oi_4 a2bb2o_1
010100 0-3 a2bb2o_2 a2bb2o_4 a2bb2oi_1 a2bb2oi_2
4-7 a2bb2oi_4 a311o_1 a311o_2 a311o_4
010101 0-3 a311oi_1 a311oi_2 a311oi_4 a31o_1
4-7 a31o_2 a31o_4 a31oi_1 a31oi_2
010110 0-3 a31oi_4 a32o_1 a32o_2 a32o_4
4-7 a32oi_1 a32oi_2 a32oi_4 a41o_1
010111 0-3 a41o_2 a41o_4 a41oi_1 a41oi_2
4-7 a41oi_4 o2111a_1 o2111a_2 o2111a_4
011000 0-3 o2111ai_1 o2111ai_2 o2111ai_4 o211a_1
4-7 o211a_2 o211a_4 o211ai_1 o211ai_2
011001 0-3 o211ai_4 o21a_1 o21a_2 o21a_4
4-7 o21ai_0 o21ai_1 o21ai_2 o21ai_4
011010 0-3 o21ba_1 o21ba_2 o21ba_4 o21bai_1
4-7 o21bai_2 o21bai_4 o221a_1 o221a_2
011011 0-3 o221a_4 o221ai_1 o221ai_2 o221ai_4
4-7 o22a_1 o22a_2 o22a_4 o22ai_1
011100 0-3 o22ai_2 o22ai_4 o2bb2a_1 o2bb2a_2
4-7 o2bb2a_4 o2bb2ai_1 o2bb2ai_2 o2bb2ai_4
011101 0-3 o311a_1 o311a_2 o311a_4 o311ai_0
4-7 o311ai_1 o311ai_2 o311ai_4 o31a_1
011110 0-3 o31a_2 o31a_4 o31ai_1 o31ai_2
4-7 o31ai_4 o32a_1 o32a_2 o32a_4
011111 0-3 o32ai_1 o32ai_2 o32ai_4 o41a_1
4-7 o41a_2 o41a_4 o41ai_1 o41ai_2
100000 0-3 o41ai_4 maj3_1 maj3_2 maj3_4
4-7 mux2_1 mux2_2 mux2_4 mux2_8
100001 0-3 mux2i_1 mux2i_2 mux2i_4 mux4_1
4-7 mux4_2 mux4_4 ha_1.c ha_1.s
100010 0-3 ha_2.c ha_2.s ha_4.c ha_4.s
4-7 fa_1.c fa_1.s fa_2.c fa_2.s
100011 0-3 fa_4.c fa_4.s fah_1.c fah_1.s
4-7 fahcin_1.c fahcin_1.s fahcon_1.c fahcon_1.s
100100 0-3 dlxtp_1 dlxbp_1.q dlxbp_1.n dlxtn_1
4-7 dlxtn_2 dlxtn_4 dlxbn_1.q dlxbn_1.n
100101 0-3 dlxbn_2.q dlxbn_2.n dlrtp_1 dlrtp_2
4-7 dlrtp_4 dlrbp_1.q dlrbp_1.n dlrbp_2.q
100110 0-3 dlrbp_2.n dlrtn_1 dlrtn_2 dlrtn_4
4-7 dlrbn_1.q dlrbn_1.n dlrbn_2.q dlrbn_2.n
100111 0-3 dfxtp_1 dfxtp_2 dfxtp_4 dfxbp_1.q
4-7 dfxbp_1.n dfxbp_2.q dfxbp_2.n dfrtp_1
101000 0-3 dfrtp_2 dfrtp_4 dfrbp_1.q dfrbp_1.n
4-7 dfrbp_2.q dfrbp_2.n dfrtn_1 dfstp_1
101001 0-3 dfstp_2 dfstp_4 dfsbp_1.q dfsbp_1.n
4-7 dfsbp_2.q dfsbp_2.n dfbbp_1.q dfbbp_1.n
101010 0-3 dfbbn_1.q dfbbn_1.n dfbbn_2.q dfbbn_2.n
4-7 edfxtp_1 edfxbp_1.q edfxbp_1.n sdfxtp_1
101011 0-3 sdfxtp_2 sdfxtp_4 sdfxbp_1.q sdfxbp_1.n
4-7 sdfxbp_2.q sdfxbp_2.n sdfrtp_1 sdfrtp_2
101100 0-3 sdfrtp_4 sdfrbp_1.q sdfrbp_1.n sdfrbp_2.q
4-7 sdfrbp_2.n sdfrtn_1 sdfstp_1 sdfstp_2
101101 0-3 sdfstp_4 sdfsbp_1.q sdfsbp_1.n sdfsbp_2.q
4-7 sdfsbp_2.n sdfbbp_1.q sdfbbp_1.n sdfbbn_1.q
101110 0-3 sdfbbn_1.n sdfbbn_2.q sdfbbn_2.n sedfxtp_1
4-7 sedfxtp_2 sedfxtp_4 sedfxbp_1.q sedfxbp_1.n
101111 0-3 sedfxbp_2.q sedfxbp_2.n ebufn_1/_2 ebufn_4/_8
4-7 einvp_1/n_0 einvp_1/n_1 einvp_2/n_2 einvp_4/n_4
110000 0-3 einvp_8/n_8 dg~sd1_1 dg~4sd2_1 dg~4sd3_1
4-7 dm~6s2s_1 dm~6s4s_1 dm~6s6s_1 clkbuf_1
110001 0-3 clkbuf_2 clkbuf_4 clkbuf_8 clkbuf_16
4-7 clkinv_1 clkinv_2 clkinv_4 clkinv_8
110010 0-3 clkinv_16 clkinvlp_2 clkinvlp_4 cdb~4s15_1
4-7 cdb~4s15_2 cdb~4s18_1 cdb~4s18_2 cdb~4s25_1
110011 0-3 cbd~4s25_2 cdb~4s50_1 cdb~4s50_2 dlclkp_1
4-7 dlclkp_2 dlclkp_4 sdlclkp_1 sdlclkp_2
110100 0-3 sdlclkp_4 lpfii~0p_1 lpfii~0n_1 lpfii~1p_1
4-7 lpfii~1n_1 lpfii~latch_1 lpfibs~_1 lpfibs~_2
110101 0-3 lpfibs~_4 lpfibs~_8 lpfibs~_16

where dg~ = dlygate, dm~ = dlymetal, cdb~ = clkdlybuf, lpfii~ = lpflow_inputiso, lpfibs~ = lpflow_isobufsrc.

The design also contains an experimental timing circuit for measuring the switching times of the individual standard cells using a ring oscillator. This is complicated by (1) a ring oscillator built from standard cells being necessarily slower than the time to be measured, and (2) several buffering and multiplexing cells plus wires also included in the measurement.

To offset (1), we can repeat a measurement several times and average them. Since the ring oscillator is not synchronized to the rest of the chip, this should result in higher timing resolution. To combat (2), we can compare the results to gate-level simulations and finetune the models until the results match up.

How to test

Set pin 1 high to switch to page mode. Find the standard cell you would like to test in the table above and set pins 2-7 to the 6 bit binary page number indicated in the first column. If pin 0 is not connected to the clock, manually toggle it low and then high to force a clock cycle. Set pin 1 low to switch to input mode. Set pins 2 and up to the values that should be supplied to the selected standard cell's input pins. Once again, you may need to manually trigger a clock cycle. The result should appear on the output pin corresponding to the table column.

To use the experimental timing circuit, make sure pin 0 is in manual mode (not connected to a clock). First set the page number the same way as above. While still in page mode, set pins 2-7 to the virtual page number 111pqr where pqr is the cell index within the page. Trigger another clock cycle using pin 0. Now set pin 1 low to switch to input mode. Set pins 2 and up to the initial cell input values and toggle pin 0 low and high again. Set pins 2 and up to the modified cell input values and toggle pin 0 low and high once more. This will latch the standard cell's previous output (i.e. the one for the initial input) and will connect the ring oscillator to a counter while the output is the same as the latched value. The counter is connected to output pins 0-5. If the cell output for the initial and modified inputs are different, this should settle to a value based on the cell switching time. Otherwise it will keep running indefinitely. Pin 6 is connected to the same gated clock as the counter but through a massive clock divider, resulting in visible blinking if the counter is still running. The blinking speed can also be measured to calculate the frequency of the ring oscillator (which depends on temperature, voltage and process parameters). Pin 7 shows the latched cell output to help debugging.

Picture

IO

#InputOutput
0clockoutput[8*page+0] / counter[0]
1page modeoutput[8*page+1] / counter[1]
2input[0] / page[0] / cell[0]output[8*page+2] / counter[2]
3input[1] / page[1] / cell[1]output[8*page+3] / counter[3]
4input[2] / page[2] / cell[2]output[8*page+4] / counter[4]
5input[3] / page[3] / 1 (timing)output[8*page+5] / counter[5]
6input[4] / page[4] / 1 (timing)output[8*page+6] / strobe
7input[5] / page[5] / 1 (timing)output[8*page+7] / latched value