@Tim Edwards For GF180MCU models available here There is subckt for nfet_03v3_dss But there is no subckt for nfet_03v3 Why? The model file intro mentions: * nfet_03v3 Subcircuit model for 3.3V NMOS But actually it is a plain model with binning: nfet_03v3.0, nfet_03v3.1, etc. Also do you plan to fix the multiplier bug we discussed before? PS: The subckt is a must to fix the multiplier issue @Stefan Schippers this may be relevant to xschem as well

@Hesham Omran It's subcircuit.

@Amro Tork • nfet_03v3_dss subckt is defined 6 times for all corners • nfet_03v3 subckt is defined one time only in fets_mm • nfet_03v3 subckt refers to the nfet_03v3 model (both have the same name) causing circular dependency errors

@Hesham Omran Ignore nfet_03v3_dss

@Amro Tork • Using the same name twice is not a good practice as it doesn't work for simulators that ignore the first letter rule like spectre

@Hesham Omran • These models are made for ngspice not for spectre. • _dss is a different device. •

Typical

lib has

fet_mm

as well and the parameters needed to complete the model:

.LIB typical
 .lib 'sm141064.ngspice' nfet_03v3_t
 .lib 'sm141064.ngspice' pfet_03v3_t
*
 .param rsh_nplus_u_m=60
 .param rsh_pplus_u_m=185
 .param nfet_06v0_vsat = 1         
 .param nfet_06v0_vth0 = 0              
 .param nfet_06v0_xl = 0
 .param nfet_06v0_xw = 0
 .param nfet_06v0_tox = 0
 .param nfet_06v0_cgso = 1
 .param nfet_06v0_cgdo = 1
 .param nfet_06v0_nvt_u0 = '0.070102'
 .param nfet_06v0_nvt_vth0 = '-0.039'
 .param nfet_06v0_nvt_xl = '0'
 .param nfet_06v0_nvt_xw = '0'
 .param nfet_06v0_nvt_tox = '1.52e-008'
 .param nfet_06v0_nvt_cgso = '1e-010'
 .param nfet_06v0_nvt_cgdo = '1e-010'
 .param pfet_06v0_dvth0 = 0            
 .param pfet_06v0_dxl = 0              
 .param pfet_06v0_dxw = 0              
 .param pfet_06v0_dtox = 0             
 .param pfet_06v0_dcgdo = 1            
 .param pfet_06v0_dcgso = 1            

 .lib 'sm141064.ngspice' nfet_06v0_t
 .lib 'sm141064.ngspice' pfet_06v0_t
 .lib 'sm141064.ngspice' nfet_06v0_nvt_t
 .lib 'sm141064.ngspice' noise_corner
 .lib 'sm141064.ngspice' fets_mm
.ENDL

Thanks @Amro Tork But I think the multiplier MC bug exists for ngspice/xschem https://www.linkedin.com/pulse/why-using-device-multiplier-can-make-your-monte-carlo-hesham-omran/

@Hesham Omran ‘Ldif’ is a user defined parameter. I should have removed it from the example. In xschem, it replaces that with a constant number to represent diffusion dimension.

@Amro Tork give the article a 5 min read and you will get it 🙂

Using m parameter is not a good move for mismatch analysis, since the '`m`' parallel devices are 100% correlated (simulator just multiplies currrents). Much better to instantiate a vectored instance, like

M1[3:0]

instead of

M1

with

m=4

. Vectored instances will just be unrolled as regular independent parallel devices in netlist,

M1[3], M1[2], M1[1], M1[0]

. In sky130 the

m

parameter is also aliased as

mult

. mult is used in addition to m since mult as a user parameter is accessible from inside the model. Standard deviation of key model parameters is scaled as 1/sqrt(W * L * mult):

delvto = {0.004+sw_vth0_sky130_fd_pr__pfet_g5v0d10v5*1.25+sw_vth0_sky130_fd_pr__pfet_g5v0d10v5_mc*1.25+sw_mm_vth0_sky130_fd_pr__pfet_g5v0d10v5*mismatch_factor*MC_MM_SWITCH*AGAUSS(0,1.0,1)/sqrt(l*w*mult)}

I tried a mismatch simulation (1000 runs) of the vth@100nA of m=10 w=0.5 l=0.15 nfet_01v8 transistor and 10 individual w=0.5 l=0.15 transistors (M1[9:0]) in parallel: there is an offset in the vth distributions.

@Stefan Schippers Exactly! That's what my article is all about. It is funny that this bug exists in many commercial PDKs.